Practical Analysis of Flavor and Fragrance Materials

Practical Analysis of Flavor and Fragrance Materials Practical Analysis of Flavor and Fragrance Materials Edited by Ke...

Author: Kevin Goodner | Russell Rouseff

236 downloads 1834 Views 5MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Practical Analysis of Flavor and Fragrance Materials

Practical Analysis of Flavor and Fragrance Materials Edited by Kevin Goodner and Russell Rouseff

A John Wiley & Sons, Ltd., Publication

This edition first published 2011 © 2011 Blackwell Publishing Ltd. Registered office John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, United Kingdom For details of our global editorial offices, for customer services and for information about how to apply for permission to reuse the copyright material in this book please see our website at www.wiley.com. The right of the author to be identified as the author of this work has been asserted in accordance with the Copyright, Designs and Patents Act 1988. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by the UK Copyright, Designs and Patents Act 1988, without the prior permission of the publisher. Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic books. Designations used by companies to distinguish their products are often claimed as trademarks. All brand names and product names used in this book are trade names, service marks, trademarks or registered trademarks of their respective owners. The publisher is not associated with any product or vendor mentioned in this book. This publication is designed to provide accurate and authoritative information in regard to the subject matter covered. It is sold on the understanding that the publisher is not engaged in rendering professional services. If professional advice or other expert assistance is required, the services of a competent professional should be sought. The publisher and the author make no representations or warranties with respect to the accuracy or completeness of the contents of this work and specifically disclaim all warranties, including without limitation any implied warranties of fitness for a particular purpose. This work is sold with the understanding that the publisher is not engaged in rendering professional services. The advice and strategies contained herein may not be suitable for every situation. In view of ongoing research, equipment modifications, changes in governmental regulations, and the constant flow of information relating to the use of experimental reagents, equipment, and devices, the reader is urged to review and evaluate the information provided in the package insert or instructions for each chemical, piece of equipment, reagent, or device for, among other things, any changes in the instructions or indication of usage and for added warnings and precautions. The fact that an organization or Website is referred to in this work as a citation and/or a potential source of further information does not mean that the author or the publisher endorses the information the organization or Website may provide or recommendations it may make. Further, readers should be aware that Internet Websites listed in this work may have changed or disappeared between when this work was written and when it is read. No warranty may be created or extended by any promotional statements for this work. Neither the publisher nor the author shall be liable for any damages arising herefrom. Library of Congress Cataloging-in-Publication Data Practical analysis of flavor and fragrance materials / edited by Kevin Goodner, Russell Rouseff. p. cm. Includes bibliographical references and index. ISBN 978-1-4051-3916-8 (cloth) 1. Flavor – Analysis. 2. Flavoring essences – Analysis. 3. Flavor – Biotechnology. I. Goodner, Kevin. II. Rouseff, Russell. TP418.P73 2011 664 .07–dc23 2011013226 A catalogue record for this book is available from the British Library. Print ISBN: 978-1-405-13916-8; ePDF ISBN: 978-1-444-34314-4; oBook ISBN: 978-1-444-34313-7; ePub ISBN: 978-1-119-97521-2; mobi ISBN: 978-1-119-97522-9 Typeset in 10.5/13pt Sabon by Laserwords Private Limited, Chennai, India

Contents Preface

xiii

About the Editors

xvii

List of Contributors

xix

1 Overview of Flavor and Fragrance Materials David Rowe 1.1

1.2 1.3

Flavor Aroma Chemicals 1.1.1 Nature Identical 1.1.1.1 Alcohols 1.1.1.2 Acids 1.1.1.3 Esters 1.1.1.4 Lactones 1.1.1.5 Aldehydes 1.1.1.6 Ketones 1.1.2 Heterocycles 1.1.2.1 Oxygen-containing 1.1.2.2 Nitrogen-containing 1.1.2.3 Sulfur-containing 1.1.3 Sulfur Compounds 1.1.3.1 Mercaptans 1.1.3.2 Sulfides Flavor Synthetics Natural Aroma Chemicals 1.3.1 Isolates 1.3.2 Biotechnology 1.3.3 ‘Soft Chemistry’

1 1 1 2 3 4 5 5 6 6 6 7 8 8 9 9 10 11 12 12 13

vi

CONTENTS

1.4

1.5

Fragrance Aroma Chemicals 1.4.1 Musks 1.4.2 Amber 1.4.3 Florals 1.4.4 ‘Woodies’ 1.4.5 Acetals and Nitriles Materials of Natural Origin 1.5.1 Essential Oils 1.5.1.1 Cold-pressing – Citrus Oils 1.5.1.2 Steam-distilled Oils 1.5.1.3 A Note on ‘Adulteration’ 1.5.2 Absolutes and Other Extracts Acknowledgments References

2 Sample Preparation Russell Bazemore 2.1 2.2 2.3 2.4

2.5

2.6

2.7

2.8

Introduction PDMS Static Headspace Extraction 2.3.1 Advantages and Disadvantages Dynamic Headspace Extraction 2.4.1 Advantages 2.4.2 Disadvantages Solid Phase Microextraction (SPME) 2.5.1 Research 2.5.2 Practical 2.5.3 Advantages 2.5.4 Disadvantages Stir Bar Sorptive Extraction 2.6.1 Research 2.6.2 Practical 2.6.3 Advantages 2.6.4 Disadvantages PDMS Foam and Microvial 2.7.1 PDMS Foam 2.7.2 Microvial Solvent Extraction 2.8.1 MIXXOR 2.8.2 Soxhlet Extraction

14 14 15 16 17 17 18 18 18 19 20 21 21 21 23 23 24 25 25 26 27 27 27 27 29 32 32 33 33 34 35 35 36 36 36 39 39 39

CONTENTS

2.9

2.8.3 Solvent Assisted Flavor Evaporation (SAFE) Summary References

3 Traditional Flavor and Fragrance Analysis of Raw Materials and Finished Products Russell Rouseff and Kevin Goodner 3.1 3.2

3.3

Overview Physical Attribute Evaluation 3.2.1 Color – Optical Methods 3.2.2 Turbidity 3.2.3 Water Activity 3.2.4 Moisture Content 3.2.4.1 Karl Fischer Method 3.2.4.2 Secondary Moisture Determination Methods 3.2.5 Optical Rotation 3.2.6 Specific Gravity 3.2.7 Refractive Index 3.2.8 Sugars/Soluble Solids 3.2.9 Viscosity Instrumental Analysis 3.3.1 Separation Techniques 3.3.1.1 Gas Chromatography (GC) 3.3.1.2 GC Retention Data 3.3.1.3 Standardized Retention Index Systems 3.3.1.4 GC Injection 3.3.1.5 GC Columns (Stationary Phases) 3.3.1.6 GC Detectors 3.3.2 Identification Techniques 3.3.2.1 Retention Index Approach 3.3.2.2 GC–MS 3.3.2.3 MS/MS References

4 Gas Chromatography/Olfactometry (GC/O) Kanjana Mahattanatawee and Russell Rouseff 4.1 4.2 4.3

Introduction Odor Assessors’ Selection and Training Sensory Vocabulary

vii

42 42 42

45 45 47 48 49 49 50 50 51 51 52 52 53 54 54 55 55 55 55 56 58 60 63 63 64 65 67 69 69 70 71

viii

CONTENTS

4.4 4.5 4.6

4.7 4.8

4.9

GC/Olfactometers (Sniffers) Practical Considerations Types of GC/Olfactometry 4.6.1 Dilution Analysis 4.6.2 Time Intensity 4.6.3 Detection Frequency 4.6.4 Posterior Intensity Method Sample Introduction Identification of Aroma-active Peaks 4.8.1 Standardized Retention Index Values 4.8.2 Aroma Description Matching 4.8.3 MS Identifications 4.8.4 Use of Authentic Standards Conclusion References

5 Multivariate Techniques Vanessa Kinton 5.1 5.2 5.3 5.4

5.5 5.6

Introduction Hierarchical Cluster Analysis (HCA) Principal Component Analysis (PCA) Classification Models 5.4.1 k-Nearest Neighbors (k-NN) 5.4.2 Soft Independent Modeling of Class Analogy (SIMCA) Principal Component Regression Example of Data Analysis for Classification Models 5.6.1 Tabulating Data 5.6.2 Examining Data 5.6.3 Multivariate Exploratory Analysis 5.6.4 Creation of a Classification Model with a Training Set and Validation with a Testing Set References

6 Electronic Nose Technology and Applications Marion Bonnefille 6.1 6.2 6.3

Introduction Human Smell and Electronic Noses Techniques to Analyze Odors/Flavors 6.3.1 Sensory Panel 6.3.2 GC and GC/MS

72 73 73 73 76 79 82 83 84 84 85 85 86 86 87 91 91 97 98 99 100 100 101 102 102 103 103 106 109 111 111 112 113 113 114

CONTENTS

6.3.3 6.3.4 6.3.5

6.4 6.5

GC/Olfactometry Electronic Nose Electronic Nose Technology and Instrumentation 6.3.5.1 Architecture 6.3.5.2 Air Generator 6.3.5.3 Sampling 6.3.5.4 Detection Technologies 6.3.6 Data Treatment Tools The Main Criticisms Directed at the Electronic Nose Market and Applications 6.5.1 Application Range 6.5.2 Perfumery Compound Detection in a Fragrance 6.5.3 Cosmetic Natural Raw Materials: Characterization of Volatile Constituents of Benzoin Gum 6.5.4 Home Care Products: Identification and Quantification Using an Electronic Nose in the Perfumed Cleaner Industry 6.5.5 Pharmaceutical Products: Flavor Analysis in Liquid Oral Formulations References

7 MS/Nose Instrumentation as a Rapid QC Analytical Tool Ray Marsili 7.1 7.2 7.3 7.4 7.5

Introduction Operating Principle Advantages of MS over Solid State Sensors Using Other Sample Preparation Modes Techniques for Improving Reliability and Long-term Stability 7.5.1 Calibration Transfer Algorithms 7.5.2 Internal Standards 7.6 Two Instruments in One 7.7 Application Examples 7.8 Classification of Coffee Samples by Geographic Origin 7.9 Classification of Whiskey Samples by Brand 7.10 Future Directions: Partnering MS/Nose with GC/MS 7.11 Conclusion References

ix

114 115 115 115 117 118 121 127 134 136 136 138

139

142 146 151 155 155 157 160 160 161 161 162 163 163 164 166 168 170 170

x

CONTENTS

8 Sensory Analysis Carlos Margaria and Anne Plotto 8.1 8.2 8.3 8.4

8.5

8.6

8.7 8.8 8.9

Introduction The Purpose of Sensory Analysis Flavor Perception Sensory Analysis Techniques 8.4.1 Overall Difference Tests 8.4.1.1 Triangle Test 8.4.1.2 Duo–Trio Test 8.4.1.3 Simple Difference Test 8.4.2 Single Attribute Difference Tests 8.4.2.1 Difference from Control 8.4.2.2 Paired Comparison Test 8.4.2.3 Ranking Tests 8.4.3 Descriptive Tests 8.4.4 Affective Tests Preparation and Planning 8.5.1 Experimental Design 8.5.2 Environment 8.5.3 Sample Preparation Panel Selection 8.6.1 Trained Panels 8.6.2 Consumer Panels Conducting a Panel Expression of Results Conclusions References

9 Regulatory Issues and Flavors Analysis Robert A. Kryger 9.1 9.2

Introduction Regulatory Overview 9.2.1 History 9.2.2 Safety Regulations 9.2.3 Product Labelling Regulations 9.2.4 Fair Trade/Conformity with Established Standards 9.2.5 Flavor Types 9.2.6 Governing Authorities 9.2.7 Role of Flavor Analysis in Regulatory Conformance

173 173 174 177 178 179 180 182 182 184 184 184 185 186 188 189 189 191 191 192 193 194 195 196 197 198 201 201 202 202 204 206 208 209 211 212

CONTENTS

9.3

Index

Specific Regulatory Issues 9.3.1 Identifying the Presence of ‘Forbidden’ Substances 9.3.1.1 Heavy Metals such as Pb, As, Hg, and Cd 9.3.1.2 Pesticides 9.3.1.3 Environmental Toxins 9.3.1.4 Allergen Testing 9.3.2 Testing Whether a Product is ‘Natural’ or Meets a ‘Standard of Identity’ 9.3.3 Testing for Other Regulatory Compliance Requirements References

xi

213 213 214 214 215 216 216 219 220 223

Preface

PRACTICAL ANALYSIS OF FLAVOR AND FRAGRANCE MATERIALS Flavor is one of the most important factors in consumer purchases and long term consumption. However, flavor is not easily quantified as the factors that impact flavor are almost always trace components. So from a chemical point of view, flavor analysis is essentially trace organic analysis. The human factor is essential to understanding flavor because humans have different genetic and cultural backgrounds which will alter their perception of flavor. Therefore all flavor analysis should be guided by human sensory panels. For too many years the study of flavor was conducted by analytical chemists who measured what they could measure using traditional analytical techniques rather than quantifying those trace impact compounds which should be measured. For many years the use of human assessors (sensory analysis) was conducted without interest in determining what was producing flavor changes in products being evaluated. Because sensory panels are impractical for routine quality control purposes, most food and fragrance manufacturers have chosen a middle ground where sensory panel data is used to guide chemists as to which compounds should be monitored to maintain quality or a specific sensory profile. This book is an attempt to demonstrate how to develop this hybrid approach to flavor analysis. The few books that exist for flavor analysis have exclusively detailed either chemical analysis with sensory input or exclusively sensory analysis without regard to chemical composition.

xiv

PREFACE

This book is aimed at the practical side of analytical analyses. We attempt to produce a book as a reference book or as a primer for analytical chemists who are starting out in the flavor and fragrance industry with useful chapters on some of the major topics that someone new to the industry might encounter, including some of the basic tests one might see in the labs such as ◦ Brix, water activity, turbidity, and similar tests. David Rowe summarized much of the descriptive information from his recent book on Chemistry and Technology of Flavour and Fragrance into the first chapter. Sample preparation techniques are described by Russell Bazemore in the next chapter. It provides a detailed description of classic and cutting edge sampling techniques that ultimately determine the success of any flavor analysis. Traditional analytical techniques that have been used to measure the quality of raw flavor materials and finished products are presented next. Gas chromatography-mass spectrometry is included in this chapter as it is the most common technique employed by flavor chemists. Gas chromatography-olfactometry, GC-O, is a hybrid technique employing the separation power of high resolution gas chromatography with the particular selectivity and sensitivity of human olfaction. This chapter written by Kanjana Mahattanatawee and Russell Rouseff, covers the hardware, software, and various techniques used for GC-O, along with selected applications, and benefits. Vanessa Kinton wrote the next chapter on multivariate techniques which are commonly used for data analysis. This chapter describes the mathematical background and theory behind these techniques. The focus is to provide a basic understanding of the theory behind these mathematical approaches knowing that in practice the procedures are handled as a ‘‘black box’’. These techniques are used extensively in many areas of analysis (electronic nose, MS chemsensor, sensory analysis, etc.) and this chapter provides the basics while the other chapters provide the application examples. Chapters 5 and 6, by Marion Bonnefille and Ray Marsili respectively, employ many of the multivariate data treatments for two very different sensor types. Chapter 5 concerns the metal oxide based electronic nose while chapter 6 is on the MS-based chemical sensor. Although both techniques employ pattern recognition software from instrumental sensors to mimic human olfaction, they differ profoundly in the types and number of sensors used to obtain the data arrays. The chapter on sensory analysis by Carlos Margaria and Anne Plotto is likely to be an area in which most chemists have little familiarity. This chapter provides a wealth of practical information about conducting

PREFACE

xv

sensory panels both trained and untrained with many anecdotes from their own experience. The last chapter describes the ever changing regulations that affect flavor analysis in the industry and is written by Robert Kryger. This is an extremely important issue that is rarely taught in schools or universities. He discusses many of the basic terms and regulations as well as some of the complications in interpreting these regulations which vary from country to country. The editors hope that this compilation will benefit those scientists beginning their careers in the area of flavor. Finally, and most importantly, we wish to thank each contributor for their time and efforts they put into their respective chapters. This book was a long time in the making and we are most appreciative of individual authors for their dedication and expertise in making this book possible.

About the Editors Kevin L. Goodner received both a B.S.Ch. in Chemistry and a B.S. in Mathematics from the University of Memphis in 1992 and a Ph.d. in Analytical Chemistry from the University of Florida working with Fourier Transform Mass Spectrometry. His focus changed to flavor chemistry after a 1.5 year post-doctorial position at the University of Florida with Russell Rouseff. Kevin then worked for 9 years at the USDA Citrus and Subtropical Products Laboratory researching flavor and quality aspects of many products. In January of 2008, Kevin switched to industry at Sensus, LLC working on tea, coffee, and other products where he currently is the Director of Research and Development. Kevin has over 50 peer-reviewed publications. Russell Rouseff is a professor at the University of Florida’s Citrus Research and Education Center specializing in flavor and color chemistry. He has 35 years experience in the Florida citrus industry, first with the Department of Citrus and then the University. He has written or edited five books, 37 book chapters and over 108 referred journal articles. He has mentored scores of domestic and international students, 8 post docs and numerous visiting scientists. He has worked with aroma volatiles in fruits, coffee, wine, flowers and foliage and bitter nonvolatiles. He is a Fellow of the American Chemical Society’s Agricultural and Food Chemistry Division, recipient of the IFT’s Citrus Products Division Research and Development Award and received the American Chemical Society’s Award for the Advancement of Food and Agricultural chemistry in 2009. Hobbies include salt water reef aquariums, tennis and motor cycles.

List of Contributors Russell Bazemore, Volatile Analysis Corporation, USA Marion Bonnefille, Alpha MOS, Toulouse, France Kevin Goodner, Sensus, LLC, Hamilton, USA Vanessa Kinton, Alcohol and Tobacco Tax and Trade Bureau (TTB), Ammendale, USA Robert A. Kryger, Citrus Resources LLC, Lakeland, USA Kanjana Mahattanatawee, Department of Food Technology, Siam University, Thailand Carlos Margaria, US Distilled Products, Princeton, USA Ray Marsili, Marsili Consulting Group,Rockford, USA Anne Plotto, USDA, ARS, Citrus and Subtropical Products Laboratory, Winterhaven, USA Russell Rouseff, IFAS, Citrus Research and Education Center, University of Florida, USA David Rowe, Riverside Aromatics Ltd , Poole, UK

1 Overview of Flavor and Fragrance Materials David Rowe Riverside Aromatics Ltd, Poole, UK

The nature of this chapter must be that of an overview as the alternative would be a multivolume series! The difficulty is not a shortage of material but rather a surfeit, and a second issue is how to give a rational coverage; should the materials be classified by chemistry, by odor or by application? The approach here is a combination of all three, and is based in part on a pr´ecis of The Chemistry and Technology of Flavours and Fragrances [1]. There is, of course, a massive overlap between flavor and fragrance; for example, cis-3-hexenol, discussed below, has a ‘green’, cut-grass odor, and hence contributes freshness to both flavors and fragrances. The division between the two Fs is itself not always a natural one!

1.1 FLAVOR AROMA CHEMICALS 1.1.1 Nature Identical The vast majority of the aroma chemicals used in flavor are nature identical (NI), that is, they have been identified as occurring in foodstuffs in the Practical Analysis of Flavor and Fragrance Materials, First Edition. Edited by Kevin Goodner and Russell Rouseff. © 2011 Blackwell Publishing Ltd. Published 2011 by Blackwell Publishing Ltd.

2

PRACTICAL ANALYSIS OF FLAVOR AND FRAGRANCE MATERIALS

human food chain. This is a key method of identifying the most important components which create a flavor, and until recently, there were also regulatory implications. European Council Directive 88/388/EEC defined these as ‘‘flavouring substances identical to natural substances’’, with the alternative being ‘‘artificial flavouring substances’’, with the latter leading to the stigma of ‘‘artificial flavors’’. The newest regulations, REGULATION (EC) No 1334/2008 OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL, no longer differentiates between Nature Identical and artificial, but the concept is still important – as a guide to flavorists, knowing a material is NI is important, and it can be especially so the context of ‘‘from the named food’’ type of flavours. Regulation 1334/2008 now only differentiates between ‘‘flavouring substances’’ and ‘‘natural flavouring substances’’, which harmonizes to an extent with the USA, where the NI classification has never been used. Even there, though, the NI concept has value, as materials have to be on the FEMA GRAS list, that is they are ‘‘Generally Recognized As Safe’’, and the vast majority of such substances are found in Nature. 1.1.1.1 Alcohols It should be noted that ethanol 1 itself is a flavor component of ‘alcoholic drinks’ as anyone tasting alcohol-free drinks will report! In fact it may considered as a solvent (especially in fragrances), as a flavour substance (FEMA 2419) or an additive (E1510)! cis-3-Hexenol 2, mentioned above, is produced in nature as a ‘wound chemical’, that is, when plant tissue is damaged, ingressing oxygen is ‘mopped up’ by reaction with linoleic acid, which generates the unstable cis-3-hexenal, which is enzymatically reduced to the alcohol. Also formed are trans-2-hexenal 3, which has a harsher, more acrid greenness and trans-2-hexenol 4, which is rather sweeter: H OH

OH 1

2

O 3

OH 4

1-Octen-3-ol, ‘mushroom alcohol’ 5, has the earthy note characteristic of mushrooms. The ‘terpenoid’ alcohols, C10 derivatives, include geraniol 6 and its isomer nerol 7, citronellol 8 and linalool 9 [2]. Cyclic terpenoid alcohols include α-terpineol 10 and menthol 11:

3

OVERVIEW OF FLAVOR AND FRAGRANCE MATERIALS

OH OH 5

6

OH OH

7

OH

8

9

OH OH 11

10

Benzyl alcohol 12 has relatively little odor and is more commonly used as a solvent in flavors; phenethyl alcohol 13 is a component of rose oil and has a pleasant rose-like aroma. Two important phenols are thymol 14 and eugenol 15, which are also major components of thyme and clove oils respectively: OH O OH OH

OH

12

13

14

15

1.1.1.2 Acids Simple acids contribute sharp notes which often become fruity on dilution. Butyric acid 16 is indisputably ‘baby vomit’ in high concentration; valeric acid 17 is cheesy, whereas 2-methylbutyric acid 18 is fruitier. Longer chain acids such as decanoic 19 are fatty and are important in

4


dairy flavors. 4-Methyloctanoic acid 20 has the sharp fatty character of roasted lamb: O

O

O OH

OH

OH

16

17

18

O

O OH

OH

19

20

1.1.1.3 Esters Numerous esters are used in flavors, so it is almost a case of any flavor alcohol combined with any flavor acid! Important simple esters include the fruity ethyl butyrate 21 and 2-methylbutyrate 22; allyl hexanoate 23 has a familiar pineapple aroma and isoamyl acetate 24 is ‘pear drops’. Phenethyl 2-methylbutyrate 25 is ‘rose bud ester’ and the warm sweet aroma of methyl cinnamate 26 makes it valuable in strawberry flavors. Methyl salicylate is the main component of wintergreen oil 27 and methyl N-methylanthranilate 28 is found in mandarin, which differentiates this from the other citrus oils: O

O

O O

O

O

21

22

23 O O

O

O O

O 24

25

26

OH

NH O

O 27

O O 28

5


1.1.1.4 Lactones These cyclic esters are usually found as gamma-lactones (five-membered rings) and delta-lactones (six-membered). Like their acyclic cousins they are used in fruit flavors and also for dairy, especially the delta-lactones such as delta-decalactone 29. Gamma-nonalactone 30, also misleadingly known as Aldehyde C18, has a powerful coconut odor:

O

O

O 29

O 30

1.1.1.5 Aldehydes Acetaldehyde 31 is ubiquitous in fruit aromas, though its volatility (b.p. 19 ◦ C) makes it difficult and dangerous to handle as a pure aroma chemical. Unsaturated aldehydes such as the previously mentioned trans-2-hexenal (leaf aldehyde) 3 are very important. trans-2-trans-4Decadienal 32 is intensely ‘fatty-citrus’; trans-2-cis-6-nonadienal 33 is ‘violet leaf aldehyde’. ‘Citral’, a mixture of the isomers geranial 34 and neral 35, is intensely lemon; it is a key flavor component of lemon and to a lesser extent other citrus oils: H

H O

O

31

3 H

O O

H 33

32 H O 34

35

H

O

Benzaldehyde 36 is widely used in fruit flavors, especially for cherry, though in fact it is not a key component of cherries. Cinnamaldehyde 37 is found in cassia and cinnamon oils. The most important aromatic

6


aldehyde, and one of the most significant of all aroma chemicals, is vanillin 38: H H

H O O

O HO O

36

37

38

1.1.1.6 Ketones The C4 ketones diacetyl 39 and acetoin 40 are used in butter-type flavors for margarines and other dairy products and hence are used in very large quantities. The former is very volatile and is believed to have led to respiratory damage amongst people exposed to large quantities of its vapor. The cyclic diketone ‘maple lactone’ 41 occurs as the enolic methylcyclopentenolone (MCP) and has the characteristic sweet, caramel odour of maple syrup. Raspberry ketone 42 is unusual in the bizarre world of flavor and fragrance trade names in that it is actually found in raspberries, tastes of raspberries and is a ketone! O

O

O OH O 39

OH 40

O 41

HO 42

1.1.2 Heterocycles [3] 1.1.2.1 Oxygen-containing The importance of materials containing the five-membered furan ring cannot be overstated [4]. Furfural 43 is formed by the Maillard reaction from pentoses in the cooking process, and 5-methylfurfural 44 from hexoses similarly. The latter has an almond, ‘marzipan’ aroma similar to benzaldehyde but with more naturalistic character. Methyl tetrahydrofuranone, ‘coffee furanone’ 45, is sweet and caramelic, but the most important flavor furan must be 2,5-dimethyl-4-hydroxy-3[2H]-furanone

7


46, an aroma chemical of many names, including strawberry furanone, and pineapple ketone. This has sweet, fruit and caramel notes, making it of obvious importance in fruit flavors, but it is also important in meat flavors, where it seems to function as a flavor enhancer. Its homologue Soy Furanone 47 is also very sweet, whereas its isomer ‘Sotolone’, or fenugreek lactone 48, has an intense fenugreek tonality, becoming more caramel-like in high dilution. The saturated furan Theaspiran 49 is found in black tea and a number of fruits: O O

O O

O H

H 43

HO

O

O

44

HO

O

45

O

OH

O

46

O

O

47

O

48

49

The most important pyrans must be Maltol 50 and Coumarin 51. The former is another caramel compound, with the latter having sweet and spicy notes. The saturated furan 1,8-cineole, or eucalyptol 52, is the main component of eucalyptus oil as well as being widespread in other oils such as lavender, distilled lime and rosemary: O OH

O O

O

O 50

51

52

1.1.2.2 Nitrogen-containing The pyrrole group is relatively unimportant in flavors, though mention should be made of 2-acetyldihydropyrrole 53, which has the ‘Holy Grail’

8


aroma of freshly baked bread but is too unstable for commercial use. The most important nitrogenous heterocycles are pyrazines, which are readily formed in the Maillard reaction from amino acids and sugars; simple alkyl pyrazines such as 2,3,5-trimethylpyrazine 54 have roasted, cocoa-like notes making them important for chocolate and roasted notes. Tetrahydroquinoxaline 55 has particularly noticeable roasted notes. 2-Acetylpyrazine 56 has very pervasive roasted, biscuit notes. The alkoxyalkylpyrazines are also found in fresh fruits and vegetables, the intensely odorous 2-methoxy-3-isobutylpyrazine 57 often being known as ‘Bell Pepper Pyrazine’. N

N

N

N

N

N

O

O

O N N

H

53

54

N

55

57

56

1.1.2.3 Sulfur-containing Whilst a few simple thiophenes are used, the most important sulfur heterocycles are thiazoles, especially 2-isopropyl-4-methylthiazole 58 and 2-isobutylthiazole 59, which have peach/tropical and tomato vine character respectively. 4-Methyl-5-thiazoleethanol, Sulfurol 60, is widely used in dairy and savory flavors. N

N

N

S

S

S

58

59

60

OH

1.1.3 Sulfur Compounds [5] The importance of sulfur compounds reflects their highly odorous character; the most odorous compounds known are sulfur compounds, with odor thresholds down to the 10−4 parts per billion level. They are the single largest group of ‘High Impact Aroma Chemicals’, materials which provide ‘character impact’ even at very low levels [6].

9


1.1.3.1 Mercaptans These are generally the most odorous of the most odorous, as it were, the capo di capi of the flavor industry. Methyl mercaptan 61 is widespread in meat aromas, as is 2-methyl-3-furanthiol (MFT) 62; the latter is especially important in beef. Furfuryl mercaptan 63 is a character impact aroma chemical of roasted coffee. The latter two are Maillard reaction products formed from cysteine and pentoses. ‘Fruity’ mercaptans include the blackcurrant/cassis materials 64 and thiomenthone 65, and p-menthene-8-thiol, the Grapefruit Mercaptan 66. The accurately named Cat Ketone, 4-mercapto-4-methyl-2-pentanone 67, is also found in grapefruit and wines. SH H3C

SH

SH

61

O

O

O

SH

62

63

64

O

O SH

SH

SH

65

66

67

1.1.3.2 Sulfides The simplest sulfide, dimethyl sulfide (DMS) 68 has a vegetable, sweetcorn odor; sulfides are less odorous than mercaptans, and hence a key aspect of quality is the need to remove all traces of mercaptans; impure DMS is quite repellent. Propyl and allyl sulfides are perhaps the commonest, especially as di- and higher sulfides; allyl disulfide 69 is the major component of garlic oil, with the remainder being mostly higher sulfides. Propyl compounds such as dipropyl trisulfide 70 are found in onion; ethyl compounds are found in Durian fruit, and to human noses other than those raised with the fruit, are at best unpleasant and sewer-like:

10

PRACTICAL ANALYSIS OF FLAVOR AND FRAGRANCE MATERIALS S

H 3C

S

CH3

S

S

68

S S

69

70

Some mercaptans oxidize very easily to form disulfides, such as the formation of bis(2-methyl-3-furyl) disulfide 71 from MFT 62: SH

S

S

O

O 62

O 71

There are a number of fruity sulfides, often derived in some way from C6 units with an oxygen atom in the 3-position relative to the sulfur; such a grouping is found in the ‘tropicals’ Tropathiane 72 and 3-methylthiohexanol 73 as well as the potato-like methional 74:

S

S

O

S

H

OH 72

O

73

74

1.2 FLAVOR SYNTHETICS There are still a number of important flavor materials which have, to date, not been found in nature. They are often used because they have properties which suitable NI materials lack; for example the so-called Ethyl vanillin 75 has a lower odor threshold than vanillin and is more soluble in organic solvents, making it more suitable for use in oil-based flavors, and ‘Ethyl maltol’ 76 is more powerful than maltol. Several glycidate esters are used, such as ethyl 3-methyl-3-phenylglycidate 77, so-called Aldehyde C16, which has a powerful strawberry aroma and is used in flavours as well as fragrances. H

O O

O OH

O O

O

O

OH 75

76

77


11

Synthetics have proved especially valuable in the area of what might be termed ‘sensates’, the molecules of taste and sensation [7]. For example, the carboxamides 78 and 79 are both cooling agents which are longer-lasting than menthol 81:

NH

NH O

O 78

79

Over the years a number of ‘synthetics’ have been found in nature, changing their status, such as the previously mentioned cat ketone and allyl hexanoate, and other materials are surely ‘synthetic’ simply because they are still hiding in foodstuffs! In addition, the ‘‘new’’ European regulations, EC 1334/2008, removes the ‘‘artificial’’ classification, and gives, potentially, a new lease of life to these materials.

1.3 NATURAL AROMA CHEMICALS [8] The seemingly innocent term ‘natural’ is, in fact, a more troublesome one than it seems. In essence ‘natural’ materials are those which are: (a) obtained by physical means from materials in the human food chain, that is, isolates; (b) obtained by biological conversions of natural materials, that is, biotechnology; or (c) obtained by reacting natural materials together in the absence of chemical reagents or catalysts, that is, cooking chemistry or soft chemistry. These definitions are enshrined in US (CFR 21, 101.22 (a) (3)) and European (REGULATION (EC) No 1334/2008) regulations. As far as aroma chemicals are concerning, ‘‘Natural’’ is a marketing conceit; the marketing departments of flavour and food companies, the supermarkets and other major retailers are unlikely to reverse their policies of promoting their subliminal (and sometimes not so subliminal) formula of Natural = Healthy, especially with the so-called ‘‘Clean Label’’ concept. The importance of the regulations is that they set the criteria which enable a material to be called ‘‘Natural’’.

12


1.3.1 Isolates A number of essential oils consisting of high levels of valuable components, make their direct isolation by physical means commercially viable. Examples include citral 34, 35 from litsea cubeba oil, anethole 80 from star anise oil, methyl N-methyl anthranilate 28 from mandarin petitgrain oil, linalool 9 from ho wood oil, and L-menthol 81 from mint oils. In the latter case, isolation is by cooling the crude mint oil to deposit the familiar large ‘bright crystals’ of commerce:

NH

H

O

O H 34

O

35

O

O

80

28

OH

OH 9

81

If the value of the end product is sufficiently high and the raw material is cheap enough, then isolation is viable even if the component is at a low level, for example valencene 82 from orange oils and nootketone 83 from grapefruit oil:

O 82

83

1.3.2 Biotechnology This very modern term actually covers one of our species’ oldest hobbies – brewing! Alcoholic fermentation, as well as forming ethanol,

13


produces other alcohols such as isobutyl 84 and isoamyl 85, 86. The latter is the main component of fusel oil, the residue after the distillation of liquors such as brandy, which can also be a source of pyrazines such as 87. N

OH OH

OH

N 84

85

86

87

Acetobacter species, unwanted contaminents in brewing, produce carboxylic acids. Gamma-lactones such as γ-decalactone 88 can be produced from ricinoleic acid 89, a component of castor oil, and vanillin 38 can be obtained from ferulic acid 90, a by-product of cereal production: OH OH

O 89

O

O 88 H

OH

O

O HO

HO

O

O 90

38

1.3.3 ‘Soft Chemistry’ This is best illustrated by the formation of esters by heating an alcohol with an acid; if the alcohol has a high boiling point this often takes place rapidly even though no catalyst is permitted. Another important

14


example is the synthesis of 2,5-dimethyl-4-hydroxy-3[2H]-furanone 46 from the hexose rhamnose 91: O

HO

OH

OH

HO

O

O

OH 91

46

This is the area of greatest contention in the natural aroma chemicals arena. EC 1334/2008 defines the techniques that may be used as ‘‘listing of traditional food preparation processes in Annex II, and does not involve, inter alia, the use of singlet oxygen, ozone, inorganic catalysts, metal catalysts, organometallic reagents and/or UV radiation.’’ but it is often easier to define what cannot be done than what can be: for example, when is a solvent also a reagent? All solvents interact with their solutes – they would not act a solvents otherwise!

1.4 FRAGRANCE AROMA CHEMICALS [9–11] As noted above, there are many aroma chemicals whose use overlaps both flavor and fragrance; many esters, aldehydes, heterocycles and indeed anything other than the most savory of flavor chemicals, tend to have uses in the fragrance sector. However, the freedom to move away from naturally occurring materials opens up a range of what we might call designer synthetics for the key notes of fragrance.

1.4.1 Musks [12] The main odiferous component of natural musk is the macrocycle Muscone 92. The scarcity of the natural material and the difficulty of synthesizing large carbocycles has driven chemists to develop synthetics since the late nineteenth century. The first artificial musks were the nitro musks such as Musk Ketone 93, discovered serendipitously during explosives research. Discoloration problems and toxicity issues have restricted the use of the nitromusks and from the 1950s the so-called polycyclic musks were introduced. These have the advantages of stability, especially in household use, as well as ease of synthesis. Galaxolide 94 is the most important of these, with other related materials including Tonalid (Fixolide) 95 and Celestolide 96. The stability and hydrophobicity of

15


these materials has led to high usage in fabric softeners and detergents as well as fine fragrances. O

O O2N

NO2

92

93

O O O 94

95

96

Advances in synthetic methods have increased the availability of macrocyclic musks. The lactone ring was relatively easy to form, and 15-pentadecanolide 97 has been sold under many names for many years; the corresponding ketone 98 is now available. Modern musk structures such as Nirvanolide 99 combine natural musk character with low odor thresholds:

O O

97

O

O

O

98

99

1.4.2 Amber ‘Amber’ materials are so-called due to their resemblance to Ambergris, a material formed in the stomachs of whales, probably as a pathological response to damage by shelly parts of plankton. This was formerly available as a by-product of the ‘whaling industry’, but now is only

16


found occasionally on beaches (a 15 kg piece was found on a beach in Australia in 2006), making it a rare and expensive material. Ambergris consists mostly of steroidal materials, and this structure forms the basis of classic ambers such as Ambroxan 100 and Amberketal 101. More recently amber molecules lacking this steroidal structure have been produced, such as Karanal 102 and Spirambrene 103:

O O

O

O

O O

H H

H

100

101

O

102

103

1.4.3 Florals Lily of the valley, or ‘muguet’, materials include the aldehydes Lyral 104 and Lilial 105. Methyl dihydrojasmonate 106 is a powerful jasmine molecule, first used in Dior’s famous ‘Eau Sauvage’. For ‘cheap and cheerful’ jasmine-type notes for household fragrances simpler materials such as α-hexylcinnamaldehyde 107 can be used:

HO

O

H

H

O

O

O

O 104

105

O 107

H

106

17


The ionones, such as α-ionone 108, were amongst the first synthetics to be used in perfumery; these violet-type materials have since been found in nature, for example in raspberry, and are also used in flavors. The damascones, such as α-damascone 109, were identified in the oil from rosa damascena; the synthetic analogue Dynascone 110 is used in the popular perfume ‘Cool Water’ by Davidoff: O

O

O 108

109

110

1.4.4 ‘Woodies’ Materials associated with this note include Iso E Super 111 and Georgywood 112; the macrocyclic ketone Trimofix 113 combines woody and amber notes: O

O O

111

112

113

1.4.5 Acetals and Nitriles Another advantage in the use of synthetics is that stability issues can be addressed; aldehydes are prone to oxidation and Aldol-type condensation reactions, especially under the harsh conditions required for household fragrances. Acetals, such as 2-methylundecanal dimethyl acetal (Aldehyde C12 MNA DMA!) 114 and nitriles such as geranonitrile 115 have similar character to their ‘parent’ aldehydes but are much less prone to these damaging reactions:

18

PRACTICAL ANALYSIS OF FLAVOR AND FRAGRANCE MATERIALS O N O 114

115

1.5 MATERIALS OF NATURAL ORIGIN 1.5.1 Essential Oils Naturally occurring materials were, of course, the foundation of both the flavor and fragrance industry and organic chemistry. The so-called essential oils are those materials obtained from plant materials by simple physical means, especially cold pressing and steam distillation. 1.5.1.1 Cold-pressing – Citrus Oils The peels of the citrus family – which includes orange, lemon, lime, bergamot, grapefruit, tangerine, and mandarin – contain glands which release oils when crushed. Cold-pressing the peel gives a mixture of water and oil, which is simply separated. The main component of all these oils is the monoterpene hydrocarbon limonene 116, typically about 95 % in orange and grapefruit, slightly lower in lemon and lime. The character of the oil is determined by ‘trace’ components, which are also considered to be ‘markers’ for the quality of the oil, for example Nootketone 83 in grapefruit (0.1–0.4 %), citral 34, 35 in lemon (1–2 %), methyl N-methyl anthranilate 28 (0.3–0.6 %) in mandarin; it should be emphasized, however, that the quality of the oils is the sum of their parts, not a marker plus carrier!

O

116

83

19

OVERVIEW OF FLAVOR AND FRAGRANCE MATERIALS NH H O O H 34

O

O

35 28

Citrus essence oils, or phase oils, are by-products of the juice market; concentration of the juice to reduce transport costs leads to oil separating out which can be sold or further processed. Whilst the essence oils are similar in components to their ‘parent’, there are some differences in that the essence oil is nearer to being an extract of the juice. The most important is orange essence oil; this contains more volatiles such as acetaldehyde and ethyl butyrate and a higher level of the sesquiterpene valencene 82 than cold-pressed orange oil.

82

The high levels of hydrocarbons in the citrus oils leads to poor solubility in water, a particular problem for their use in soft drinks. This can be overcome, in part, by folding; the oil is ‘folded’ by distilling off the more volatile monoterpene hydrocarbons, in effect concentrating the more valuable flavor components and removing the most hydrophobic components. The distillates also have value as solvents, diluents and, since the terpenes do carry over some odiferous components, as flavor or fragrance ingredients in their own right; for example in creating a tenfold orange oil, 9 kg of orange terpenes are produced per kg of ‘Orange Oil 10X’. 1.5.1.2 Steam-distilled Oils Most plant materials contain much less volatile oil than the citrus fruits, often less than 1 %, and these are simply steam distilled to obtain the oil; examples include cassia, cinnamon, mint, rose and lavender. This process

20


can, of course, lead to changes in composition due to thermal decomposition, oxidation and hydrolysis; by tradition, the best lavender oil is produced at high altitude, as the lower boiling point of water leads to less hydrolysis of esters such as linalyl 117 and lavandulyl 118 acetates.

O O O O 117

118

Chemical changes can also be beneficial. The most important lime oil is ‘‘Distilled Lime’’; this is not redistilled cold-pressed oil but oil which is steam distilled from the macerated fruit. The high acidity of the juice leads to hydration of the hydrocarbons, giving high levels of α-terpineol 10 and 1,8-cineole 52. This oil has a fresh, juicy aroma which contrasts with the waxy, floral odour of the cold-pressed oil.

O

OH 10

52

1.5.1.3 A Note on ‘Adulteration’ Historically, especially prior to the routine availability of gas chromatography, the addition of cheap materials to ‘cut’ or adulterate oils, was more widespread. It is still a problem; without the use of sophisticated isotopic analysis, the addition of a quantity of a synthetic material, for example, cinnamaldehdye to cinnamon oil, is impossible to detect, let alone prove. Similarly, the addition of a low percentage of terpenes to a citrus oil which is already over 90 % terpenes, is not going to be detected easily. Ultimately the answer has to lie in the trust in the supplier, as applied with a modicum of common sense. Customers are themselves


21

part of the problem – the desire for lower priced raw materials is itself a driving force in adulteration. Put simply, if you wish to buy an oil at 50 % of the market price, then do not be surprised if the product you are buying is only 50 % oil!

1.5.2 Absolutes and Other Extracts This final section in fact covers the oldest approach to flavor and fragrance materials – the extraction of aromatic materials into organic solvents. Extraction of plant material with a nonpolar solvent such as hexane, followed by removal of the solvent, yields a concrete. As the name implies, this is often solid or semisolid due to the presence of plant waxes as well as pigments and nonvolatiles. Extraction into ethanol followed by solvent removal gives an absolute, which is more manageable and the more usual item of commerce. This is a very labour-intensive process and hence absolutes are more expensive than most oils, and are more associated with higher-value materials such as violet flower absolute and orange flower absolute. Oleoresins are more associated with spice oils such as ginger and garlic and are a solution of an absolute-type extract in an essential or solvent such as vegetable oil or propylene glycol.

ACKNOWLEDGMENTS Many thanks to the contributors to The Chemistry and Technology of Flavours and Fragrances, whose work I have cheerfully plundered for this chapter!

REFERENCES 1. Rowe, D. J. (ed.) (2004) The Chemistry and Technology of Flavours and Fragrances, Blackwell, Oxford, UK. 2. Teisseire, P. J. (1994) Chemistry of Fragrant Substances, VCH, New York, USA. 3. Zviely, M. (2004) Aroma chemicals II: heterocycles, in The Chemistry and Technology of Flavours and Fragrances, (ed. D. J. Rowe), Blackwell, Oxford, UK, pp. 85–115. 4. Rowe, D. (2004) Fun with Furans, Chemistry and Biodiversity, 1, 2034– 2041. 5. Jameson, S. B. (2004) Aroma chemicals III: sulfur compounds, in The Chemistry and Technology of Flavours and Fragrances, (ed. D. J. Rowe), Blackwell, Oxford, UK, pp. 116–142.

22


6. Rowe, D. J. (2002) High impact aroma chemicals, in Advances in Flavours and Fragrances: From the Sensation to the Synthesis (ed. K. Swift), Royal Society of Chemistry, Cambridge, UK, pp. 202– 236. 7. Dewis, M. L. (2004) Molecules of taste and sensation, in The Chemistry and Technology of Flavours and Fragrances, (ed. D. J. Rowe), Blackwell, Oxford, UK, pp. 199–243. 8. Margetts, J. (2004) Aroma chemicals V: natural aroma chemicals, in The Chemistry and Technology of Flavours and Fragrances, (ed. D. J. Rowe), Blackwell, Oxford, UK, pp. 169–198. 9. Frater, G., Bajgrowicz, J. A. and Kraft, P. (1998) Fragrance chemistry. Tetrahedron, 54, 7633– 7703. 10. Kraft, P., Bajgrowicz, J. A., Denis, C. and Frater, G. (2000) Angew. Chem. Int. Ed., 39, 2980– 3010. 11. Sell, C. (1999) Ingredients for the modern perfume industry, in The Chemistry of Fragrances (eds D. Pybus and C. Sell), Royal Society of Chemistry, Cambridge, UK, pp. 51–124. 12. Kraft, P. (2004) Aroma chemicals IV: musks, in The Chemistry and Technology of Flavours and Fragrances, (ed. D. J. Rowe), Blackwell, Oxford, UK, pp. 143– 168.

2 Sample Preparation Russell Bazemore Volatile Analysis Corporation, USA

2.1 INTRODUCTION While injecting samples ‘neat’ (i.e., directly into the GC with no sample preparation) is a common practice for liquid samples in the flavor and fragrance industry, the practice is straightforward and requires little to no explanation or discussion. Therefore, preparation of samples by extracting volatiles for subsequent introduction into a gas chromatograph is the focus of this chapter. New or relatively new extraction techniques including solid phase microextraction (SPME), stir bar sorptive extraction (SBSE), polydimethylsiloxane (PDMS) foam, and Microvial are discussed in addition to tried and true technologies including static/dynamic headspace, and the solvent extraction methods MIXXOR, soxhlet, and Solvent Assisted Flavor Evaporation (SAFE). The use of sample prep techniques that are rapid, precise, and may be automated have become popular and necessary in industry and academia due to an increasing number of projects managed and sample analyses required. A large portion of this review is devoted to products that incorporate polydimethylsiloxane (PDMS), a polymeric material proven to be useful for volatile extraction and fast, relatively simple Practical Analysis of Flavor and Fragrance Materials, First Edition. Edited by Kevin Goodner and Russell Rouseff. © 2011 Blackwell Publishing Ltd. Published 2011 by Blackwell Publishing Ltd.

24


procedural techniques that lend themselves to automation for multisample analysis. A few basic points regarding PDMS are helpful in understanding its unique characteristics and capabilities.

2.2 PDMS Siloxanes contain a Si-O-Si backbone, and those with organic groups attached to silica are called polyorganosiloxanes with the structure indicated in Figure 2.1. PDMS is a versatile substance. It is a primary component in Silly Putty, a substance well known to many baby-boomers as a malleable childhood toy. It is also a component in silicone grease and lubricants, defoaming agents, cosmetics, hair conditioner and a filler fluid in breast implants [1]. PDMS exhibits unique flow (rheological) properties. If placed on a given surface for several hours it will flow to cover the surface and will also cover any imperfections on the surface [2]. Polymer length and/or branches or cross links dictate viscoelasticity but in general at high temperatures PDMS resembles a very viscous liquid and at low temperatures it resembles an elastic solid. It has been used extensively as a stationary phase in gas chromatography (GC), and can be used over a broad temperature range (−20 ◦ C to 320 ◦ C, [3]). It is considered to be inert, nontoxic and nonflammable. An important characteristic of PDMS is that it is hydrophobic. It does not bind water appreciably while it does extract other volatile components present in a sample matrix (immersed in a liquid or from headspace) by absorption into the polymer liquid phase. It does not require the use of solvents. These are the principal reasons why it has become popular for extracting volatiles and semivolatiles from foods, beverages, and biological materials. The octanol–water partition coefficient (KO–W ) is the ratio of a compound’s concentration in octanol and concentration in water at equilibrium at a specified temperature. The logarithmic ratio of the concentrations of solute in solvent is the definition of log P. PDMS extraction capacities can be predicted based on octanol-water partition

Si

O

Si

O

Si n

Figure 2.1 Structure of PDMS.

SAMPLE PREPARATION

25

coefficients. Generally lipophillic substances have larger KO–W values and are more readily absorbed by PDMS [4, 5]. When PDMS is exposed to organic solvents, especially pentane and xylenes, they diffuse into the polymer causing it to swell [2]. For this reason manufacturers of the two most utilized PDMS products in sample preparation chemistry, SPME (Supelco, Bellefonte, PA, USA) ¨ and Twister (Gerstel GmbH, Mulheim an der Ruhr, Germany) recommend aqueous extractions when directly immersing the SPME fiber or Twister. When conducting headspace analysis of most organic solvents, swelling still occurs and chromatograms display large solvent peaks, thus undermining a main advantage of PDMS extraction.

2.3 STATIC HEADSPACE EXTRACTION Headspace refers to the gas phase located above a liquid and solid phase present in a sealed vessel. Volatility, or rate of evaporation of a given component, is governed by both Henry’s law constant (gas to liquid distribution constant) and vapor pressure. Volatile compounds partition into the gas phase due to a number of factors, all interrelated in their effects, including propensity for solubility in water (hydrophilic or hydrophobic), polarity, ionic nature of analyte and solvent, molecular weight, and temperature. In static headspace extraction, the liquid and/or solid sample is placed in a sealed vial with an inert septum, usually Teflon (inert to prevent volatiles from sticking to the surface via adsorption, or being absorbed into the septum material). Volatile components are allowed to reach equilibrium between liquid/solid and gas phases. An aliquot of headspace is removed, normally by gas-tight syringe, and injected directly into a gas chromatograph injection port. A good rule of thumb to optimize the rate and concentration of compounds in headspace is to fill the vial to two-thirds full with sample thus leaving a third for headspace. Too large a volume of headspace requires more time to reach equilibrium conditions. Too small a volume and there may be insufficient to extract for analysis and wasted sample.

2.3.1 Advantages and Disadvantages Static headspace extraction is simple, inexpensive, and lends itself to automation. It provides a true representation of volatile compounds responsible for aroma because it reflects natural headspace concentrations. However, this method is notorious for poor precision due to

26


nonequilibrium conditions during extraction or selective changes in volatile concentrations associated with serial extractions. It is also difficult to detect potentially important components because the volatiles are not concentrated. Only the volatiles present in greatest quantities will be detected by direct headspace sampling [5].

2.4 DYNAMIC HEADSPACE EXTRACTION Dynamic extraction, also called purge and trap, differs from static extraction in that volatile compounds are continuously swept from the headspace into a trap by flow of inert carrier gas, nitrogen or helium (and sometimes air with the risk of oxidation). Figure 2.2 is a diagram of a typical purge and trap apparatus. Once trapping is complete the volatiles are released for chromatographic analysis. Rapid heating is the most efficient method for release of components from a trap. Inert gas then carries the desorbed volatiles onto the GC column to form a tight band for optimal chromatography. Other methods release volatiles by solvent desorption followed by solvent evaporation under a stream of inert gas, and then injection into a gas chromatograph. Traps may contain one or combinations of substances. Included in the list is activated carbon, and modified carbon products with uniform pore sizes. Tenax (2,6-diphenylene-oxide polymer) is commonly used in this application due to its stability when heated (upper limit of 350 ◦ C) and hydrophobic properties. PDMS foam is a newer product suitable for purge and trap and it will be discussed further in Section 2.6. Inert gas sweeps headspace from sample

Trap Inert gas exits the trap

Sample in a water bath

Figure 2.2 Diagram of a typical dynamic headspace extraction or purge and trap apparatus.

SAMPLE PREPARATION

27

Cryocooling with liquid nitrogen (b.p. −196 ◦ C) to condense volatiles onto glass wool and/or adsorbent material enhances extraction capacity. Following completion of extraction, traps are rapidly heated to desorb volatiles, as previously described.

2.4.1 Advantages Purge and trap allows for concentrating samples and therefore this method has lower limits of detection. Solid samples may also be analyzed because the confounding equilibration variable associated with solid samples is not a concern. There is a continuous introduction and departure of carrier gas into and out of a vessel. Cryocooling increases the number of highly volatile, lower molecular weight compounds extracted.

2.4.2 Disadvantages Disadvantages of purge and trap include Tenax’s low surface area and a low adsorption capacity. The method has a propensity for extracting nonpolar components and a low affinity for polar compounds. Reineccius [5] provides a good discussion of Tenax’s adsorption capabilities and shortcomings. Due to ice buildup, aqueous samples are problematic for cryocooled traps. For this reason a means for removing water vapor from the carrier gas stream must be introduced prior to the cold trap.

2.5 SOLID PHASE MICROEXTRACTION (SPME) Arthur and Pawliszyn [6] first utilized PDMS coated on a fused silica fiber to extract analytes from an aqueous media in a process called solid phase microextraction. Diagrams of manual and auto sampler SPME fiber assemblies are indicated in Figure 2.3 and the procedure is indicated in Figure 2.4. This method was commercialized and made available by Supelco Corp. (Bellefonte, PA, USA).

2.5.1 Research Since Pawliszyn’s initial work he and many other researchers have successfully utilized SPME to investigate a plethora of topics associated with volatile and semivolatile components. Among these are Marsili [7, 8] who, in addition to a broad range of flavor work in foods and beverages,

28


For manual sampling (Fiber exposed) Plunger Barrel

For auto samplers or SPME/HPLC interface (Fiber retracted)

Plunger

“Z” Slot

Barrel

Retaining screw Plunger retaining screw Hub-viewing window Adjustable needle guide/depth gauge Fiber-attachment needle

Septum-piercing needle Coated SPME fused silica fiber

Slot Color-coded screw hub

Fiber-attachment needle Retaining nut

Sealing septum

Needle ferrule

Septum-piercing needle

(a)

Coated SPME fused silica fiber (b)

Figure 2.3 Diagrams of (a) manual and (b) auto sampler solid phase microextraction devices.

(a)

(b)

(c)

(d)

(e)

(f)

Figure 2.4 SPME diagram: (a) protective cover piercing septum; (b) fiber exposure and insertion into sample; (c) fiber withdrawn into cover and removed from sample; (d) cover inserted into GC injection port; (e) fiber exposed; (f) fiber withdrawn and removed from injection port.

has extensively investigated dairy flavors. Rouseff [9, 10] utilized SPME and published comprehensively on a diverse list of food and flavor topics with a special emphasis in citrus. Wright [11–13] identified key aroma impact components and off odors in a variety of foods, beverages and agricultural products.

29

SAMPLE PREPARATION

2.5.2 Practical

Propanal

Nitropropane

108735 15720

Acetone

311 528 2119 1671

208 121 758 499 5930

14420 857 1062 7229 3209

316 192 1306 837 7829

PDMS PAcrylate PDMS-DVB CW-DVB DVB-CAR Carboxen

59563

55870

69329

As well as the original PDMS fiber coating, a variety of coatings is now available that adds to SPME extraction capabilities and includes Carboxen, divinyl benzene, polyacrylate, and carbowax (polyethylene glycol; PEG). Coatings may be combined or available separately, or both. The manufacturer recommends 85 μm Carboxen PDMS for extracting components with 125 mw (molecular weight) or less. For larger components of interest, polarity plays a larger role in extraction efficacy. Polyethylene glycol (PEG), polyacrylate (PA), and DVB/Carboxen fibers are recommended for polar analytes [14]. Figure 2.5 shows area response by fiber type and Figure 2.6 shows analyte concentration vs. response for a Carboxen–PDMS fiber. Fibers with different film thickness are also available. A thicker film coating allows for more analyte loading into the liquid polymer coating (absorption), and by theory more analyte for analysis and detection. For example there are three fibers available with different PDMS film thicknesses: 100 μm, 30 μm, and 7 μm. The larger 100 μm film will absorb more analyte, but may also release the analyte less efficiently, especially larger molecular weight semivolatiles. This may result in carryover where peaks are seen on subsequent analyses. A 30 μm coating is

Propionitrile

Figure 2.5 Area response by fiber type. PDMS is polydimethylsiloxane; PAcrylate is polyacrylate; PDMS-DVB is polydimethylsiloxane-divinylbenzene; CW-DVB is carbowax-divinylbenzene. Source: Supelco Corp.

30

PRACTICAL ANALYSIS OF FLAVOR AND FRAGRANCE MATERIALS 6.5

Pentane Nitropropane Methylene chloride Propionitrile 15 Min Ext Acetone Isopropanol Dioxane

5

Log response

5.5 5 4.5 4 3.5

5ppm

5ppb

3 2.5 2 1.5 0.0

1

1.5

2

2.5

3

3.5

4

4.5

5

5.5

Log of concentration (PPB)

Figure 2.6 Carboxen–PDMS SPME concentration vs. analyte response. Source: Supelco Corp.

intended for analysis of semivolatiles and the 7 μm fiber is intended for immersion and extraction of semivolatiles with molecular weights of 250 or more. In addition to the original fused silica fiber, a new metal alloy fiber is available. The advantage of this new core material is its improved durability and usefulness in extractions associated with the popular CTC robotic auto sampler that has a capability to agitate samples while exposing the fiber to sample or sample headspace (CTC, Zwingen, Switzerland). Fibers also come in a variety of needle gauges. The smaller 24 gauge fiber is recommended for extractions and injections utilizing the manual holder and silica GC inlet septa. This is to decrease septa coring. The slightly larger 23 gauge fiber is recommended for use with the Merlin Microseal (Merlin Instruments, Half Moon Bay, CA, USA) septum that has a longer life (should last for 1 year or 25 000 injections) and does not contaminate injection port liners with shaved silica particles. The larger size is needed here to ensure no leakage into the injection port. Another important item to consider when conducting extractions with SPME and optimizing chromatography is the GC injection port liner. A 0.75 mm injection port liner delivers superior chromatographic results compared with the normal 2 mm internal diameter liner when using

SAMPLE PREPARATION

31

SPME. Peak shape and less tailing are noticeably improved due to less band broadening. SPME fibers must first be conditioned before initial use based on the manufacturer’s specific conditioning recommendations (a GC injection port or similar device created for fiber conditioning). This step is important because residual artifacts from manufacturing and the environment are baked off and could otherwise obscure important chromatographic peaks. Additionally fibers benefit from brief bake-outs prior to subsequent daily use to remove residual volatiles and volatiles collected from the laboratory atmosphere. It is good practice to conduct blank analyses with SPME periodically, especially if a series of extractions are conducted with an autosampler, to ensure fibers have not collected artifacts or environmental contaminants. Unlike conventional static headspace extraction where volatile partitioning between the sample and sample headspace is a limiting factor, successful SPME results depend on two separate partitioning coefficients. The first is the same as with convention static extraction, that is, partitioning between sample and sample headspace. This partitioning has a practical application in determining how much time is required for components of interest to pass from the liquid or solid sample into the sample headspace, or how long should the sample be placed in a sealed container prior to extraction. Thus, unless equilibrium and exposure times are carefully duplicated, headspace concentrations will not be the same from one analysis to the next and precision will be unacceptable. If sample components are not allowed sufficient time to reach equilibrium between the two phases (sample and headspace) the resultant analyte headspace concentration will be lower. However, most procedures do not wait for equilibrium conditions, requiring precise control of exposure time to maintain reasonable reproducibility. It should be noted there are ways of increasing headspace analyte concentrations including increasing sample temperature and changing the ionic strength of aqueous sample (add salt). Also by ensuring headspace volume is approximately one third of the volume of the sample, a maximum headspace concentration will result [15]. The second partitioning effect between sample and fiber may be considered in terms of how long the fiber is exposed to the sample. Another factor which is important in complex samples is the competitive absorption between headspace analytes. As briefly discussed previously, headspace components absorb into PDMS, or adsorb to particles embedded in PDMS, for example, Carboxen, divinyl benzene.

32


2.5.3 Advantages Because of the dual equilibrium associated with SPME functionality, semivolatiles are more easily extracted than with conventional headspace methods. This is because once most semivolatile analytes are present in the headspace they are readily extracted by a PDMS fiber. These larger analytes are retained in the fiber coating where a concentration effect occurs that improves recovery (compared with static and dynamic headspace analyses). Also, unlike traditional headspace methods that may require heating samples above 100 ◦ C, SPME will commonly extract these analytes at lower temperatures usually ranging between 45–70 ◦ C [16]. SPME is very suitable for comparing samples of the same product, determining if a product has been adulterated, or determining a source of a flavoring or fragrance. Once familiar with the basic tenets of SPME methodology, a researcher will appreciate the capacity for rapid sample analysis, simple automation, and minimal cost. Because there is no need for organic solvents, the hassle and expense associated with hazardous wastes are minimized. The opportunity for good accuracy and precision are easily attainable. Additionally, by employing innovation, unconventional volatile extractions are possible. An example of thisis novel work which was conducted by Payne et al., [17] who utilized SPME to extract volatiles directly from the oral cavity.

2.5.4 Disadvantages SPME may not be an answer to all questions regarding volatile profiles. A researcher requires an accurate picture of what is occurring regarding propensities for extractions of certain functional groups vs. others. PDMS has a greater affinity for nonpolar volatiles [4], although with the addition of Carboxen, and other fiber coating ingredients and film thicknesses, this problem is improved. Based on the discussion above, the extraction of components by PDMS may follow a model predicted by log P, or the octanol–water partition coefficient. Additionally extraction efficiency is determined by the phase ratio (β) defined by: β = (V aqueous phase/V PDMS phase) Given a 100 μm PDMS coated fiber has approximately 0.5 μL of PDMS, variable extraction for nonpolar volatiles may result due to competition between the coated fiber, the aqueous phase, and the glass walls

SAMPLE PREPARATION

33

of the sample vessel [4, 5]. Such competition on an absorbent PDMS fiber is more common if one analyte is at a much higher concentration and saturates the absorbent phase. Adsorbent type fibers are more prone to competition due to limited pore sites. An analyte with higher affinity can displace an analyte with less affinity. This problem can be greatly reduced by shortening extraction times. It is difficult to quantify complex mixtures using SPME, since it is not an exhaustive extraction technique. If quantification is required for samples with a variety of analytes, SPME is probably not the best extraction tool. However, it can be an ideal tool for quantifying a specific analyte, or a small group of analytes, especially if an internal standard is utilized. As previously stated PDMS may result in sample carryover especially with larger semivolatiles. The durability of fibers as well as precision were both once a problem but technical advances in these areas have resulted in major improvements.

2.6 STIR BAR SORPTIVE EXTRACTION Stir bar sorptive extraction (SBSE) was developed by Baltussen et al. [18] and is similar to the technology of the SPME fiber that incorporates a coating of PDMS (only). GERSTEL Twister is the commercial product that utilizes this technology and is manufactured and available from ¨ Gerstel GmbH. (Gerstel, Mulheim an der Ruhr, Germany). Twister is made of an inner, magnetized, metallic bar. A glass layer covers the metal bar and PDMS coating covers the glass (Figure 2.7). Currently there are four different sizes available that range from 10 mm length and 500 μm phase thickness with 24 μL PDMS up to a length of 20 mm, 1000 μm phase thickness and 126 μL of PDMS. The larger size of SBSE allows for a greater amount of PDMS coating and more efficient stirring of larger volumes. Twister has approximately 50–250 times more extraction phase than the comparable 100 μm PDMS coated SPME fiber and solute detection limits of less than 1 ng/L (ppt) have been reported [19].

2.6.1 Research SBSE usefulness has proved diverse and capable for extracting volatiles and semivolatiles from aqueous media, foods, and beverages. A few of the many studies include those by Novotny’s group who utilizes SBSE for extraction of aqueous media in chemical ecology and biological media [17, 18, 20]. Demyttenaere [21] measured malt whiskey, Kishimoto

34


Magnet PDMS Glass

Figure 2.7 Twister consists of a magnetic stir bar coated with a layer of glass and polydimethylsiloxane. Source: Gerstel Inc.

´ [23] extracted semivolatiles measured terpenoids in beer [22], and Leon from water. Gurbuz and Rouseff [24] successfully utilized SBSE to extract wine volatiles for analysis by gas chromatography olfactometry. David and Sandra [25] have an excellent Stir Bar Sorptive Extraction review article currently in press.

2.6.2 Practical Twister may be added directly into a liquid sample or it may be utilized for headspace extraction (sometimes referred to as HSSE). In the latter an open glass insert is available from Gerstel to suspend the Twister in the headspace above the sample. At a pinch a paperclip inserted through a septum will also work. The magnetic attraction between the stir bar and paper clip will hold the stir bar suspended above the sample. Twister, like SPME, must be preconditioned prior to use and this is accomplished by heating as per the manufacturer’s guidelines. Following an extraction, Twister is manually placed into a TDU tube, or glass desorption liner. The desorption process differs from that associated with SPME where the polymeric coated fiber is inserted into a hot GC injection port. Besides the obvious that Twister is too large for desorption in a heated injection port, the thicker PDMS coating on Twister requires additional time for desorption and requires refocusing prior to transferring to the analytical column. Gerstel has automated this process (after placing Twister into a desorption liner and putting the liner in an auto sampler storage tray). If using the MPS2 auto sampler

SAMPLE PREPARATION

35

(Gerstel) desorption liners containing Twisters are placed in the MPS2’s storage tray fitted with cylindrical glass tubes to house up to 98 Twisters (two trays or up to 196 liners may be continuously analyzed). The MPS2 auto sampler transfers liners from the storage tray to the Twister Desorption Unit (TDU) located above and attached to a cooled injection system (CIS). The CIS is located on the GC in the same position as standard on-column or split/splitless injectors. The TDU may be cooled below ambient temperature by integration with a Peltier cooling device (utilizes a combination of water–ethanol) and then heated to desorb volatiles concentrated in Twister (Twister is located inside the heated TDU). The TDU is attached to a liquid nitrogen cooled injection system (CIS) that contains a glass liner (with glass wool insert) and functions to cryorefocus (trap) components desorbed from Twister in the TDU. Following this period of cryotrapping volatiles that are transferred from the TDU, the CIS has programmable heating to thermally desorb comp onents (components that have been cryofocused in the liner located inside the CIS) by rapid vaporization. Components are then transferred in either the split or splitless mode onto the analytical column for chromatographic separation. The TDU may be cooled again by the Peltier device to decrease time between multiple sample analyses.

2.6.3 Advantages SBSE has established a good track record of performance. Twisters are durable and have a useful life of approximately 50–300 extractions depending upon extraction media (E. Pfannkoch, personal communication). As previously discussed, SBSE has more PDMS coating and thus a larger capacity for volatile extraction vs. PDMS SPME. Loughrin [26] measured extraction of volatile components from waste water and found the performance of the PDMS stir bars compared favorably with that of the polar SPME fibers at octanol–water (log kow ) values > 1.50. Bicchi [27] reported both direct (placed in liquid) and headspace SBSE extractions yielded greater volatile recoveries from Arabica coffee than obtained by SPME. Also important is the fact that this system may be automated with the Gerstel MPS2 and up to 196 samples may be analyzed continuously.

2.6.4 Disadvantages Currently SBSE (Twister) products are available only with PDMS coating. An adsorption material that would aid in binding highly volatile

36


polar components would be useful. Carryover may be a concern with the thicker PDMS coating. Bake-out times must be sufficient to desorb all volatiles/semivolatiles and occasional blank runs should be conducted to ensure there is no carryover from previous analyses. Equipment necessary to utilize Twister to its fullest, including Gerstel’s TDU, CIS, and MPS2 requires some capital investment.

2.7 PDMS FOAM AND MICROVIAL Two novel and relatively new extraction methods from Gerstel, PDMS foam and Microvial, provide increased capabilities and will be briefly discussed below.

2.7.1 PDMS Foam In addition to PDMS on a fiber (SPME) and stir bar (SBSE), it is also available as a foam plug loaded into a TDU tube (available through Gerstel). Each tube contains 85 mg +/− 2.7 mg of PDMS foam. This method offers the largest PDMS mass and surface area available and can be used to trap volatile components in a dynamic headspace extraction process. Desorption is conducted utilizing the same equipment as with Twister (Gerstel’s MPS2, thermal desorption unit and cooled injection sytem). In samples with a lot of water, vapor may condense inside the TDU tube during the purge and trap process. Formation of an ice plug in the cryocooled liner may result. A nice feature of this system is the capability to purge tubes with nitrogen and vent water vapor prior to cryorefocusing absorbed components. Marsili and Laskonis [28] constructed a dynamic extraction apparatus (Figure 2.8) that recirculated air through a PDMS foam tube. Results after 5 minutes of recirculation are displayed in Figure 2.9. They concluded this extraction technique provided the accuracy (as measured by standard calibration curves) and precision necessary for most analytes studied.

2.7.2 Microvial This procedure refers to a novel technique that evolved from problems associated with direct liquid injection of samples contaminated with nonvolatile components (a process sure to decrease column life). A disposable glass microvial capable of holding up to approximately 200 μL of liquid (more liquid than in a normal liquid injection) is

37

SAMPLE PREPARATION Air to inlet of oil-free recirculating pump

Air to inlet of oil-free recirculating pump

Teflon tubing

PDMS foam trap 10 mL beer sample (with stir bar); purge stream above sample

Water bath (room temperature)

Figure 2.8 Recirculating dynamic headspace extraction apparatus with PDMS foam as the trap. Source: Marsilli Consulting Group. Abundance

3e+007

2.5e+007

2e+007

1.5e+007

1e+007

5e+006

Time (s)

200

400

600

800

1000

1200

1400

1600

1800

Time (s)

Figure 2.9 Results from PDMS foam extraction of beer headspace utilizing recirculating extraction apparatus. Source: Marsilli Consulting & Gerstel Corp.

placed inside a TDU tube (Figure 2.10). Both are placed into the thermal desorption unit. As discussed with PDMS foam, purge and programmable heating capabilities of the thermal desorption unit allow for evaporation and venting of solvent/water and cryorefocusing volatiles

38


onto an inlet liner. The volatiles trapped in the liner are then vaporized by rapid heating and transferred onto the column without problems related to nonvolatiles and an overload of solvent. A big advantage of this method is efficiency and the means to analyze multiple samples.

Cap Microvial

TDU tube

Figure 2.10 Microvial constituents including: (a) TDU tube, (b) microvial, and (c) metal cap. The cap with visible O-rings functions to provide an airtight seal so that the TDU can maintain system pressure. It also functions as a means for the auto sampler to grip and transfer the tube between the storage tray and TDU. Source: Gerstel Inc. Abundance 120000

Diazinon

40000

Hexachlorbenzene

Carbaryl

60000

Trifluralin

80000

Methiocarb o-Phenylphenol

100000

20000

8.50

9.00

9.50 Time (min)

10.00

Figure 2.11 Microvial analysis. Overlay of 10 ppb pesticide in standard and 10 ppb pesticide in carrot juice extract. Source: Gerstel Inc.

SAMPLE PREPARATION

39

Gerstel has automated it entirely from injection of a dirty sample into the microvial and transfer to the thermal desorption unit to desorption, analysis, and removal of microvial. Pfannkoch, Whitecavage, and Stuff [29] analyzed by microvial 10 μL of carrot extract spiked with 10 ppb pesticides and the same concentration of pesticide in solvent (Figure 2.11). The peak sizes of pesticides obtained from the complex carrot juice sample were similar in size and shape to those from the much cleaner standard sample. After analysis, the microvial and residual nonvolatile material from the extract were disposed of, minimizing contamination of the inlet.

2.8 SOLVENT EXTRACTION In its simplest form solvent extraction is conducted by the addition of a liquid or solid sample to a liquid solvent, agitation to ensure mixture, and removal of the solvent plus that which dissolved in it. However, many factors affect solubility and include lipophilicity, polarity, temperature, pH, molecular weight, pressure, mixing thoroughness, and time. A brief overview of some liquid extraction methods follow.

2.8.1 MIXXOR A separatory funnel is probably the best known solvent extraction tool. The sample is added to immiscible liquids, shaken, and the appropriate layer with solute is removed. MIXXOR (NBS Systems, Haifa, Israel) improves separation due to vigorous blending and corresponding mass transfer (Figure 2.12). Samples processed with MIXXOR or a similar device, and centrifuged to separate layers, have provided successful results. Parliament [30] reported success utilizing a similar apparatus and Jella et al. [31] followed this method to extract flavor components from grapefruit juice.

2.8.2 Soxhlet Extraction A liquid extraction process for utilizing solvent to extract solid samples involves the Soxhlet extractor (Figure 2.13). A container (thimble) open at the top, made of permeable, filter paper-like cellulose, is placed inside the soxhlet extractor and positioned directly beneath a condenser. Solvent is added to a round bottom flask and heated with a heating mantle.

40


Figure 2.12 A MIXXOR apparatus takes the separatory funnel concept a step further in efficiency. Source: Sigma Aldrich.

Condenser

Soxhlet extractor Thimble

Round bottomed flask

Figure 2.13 Soxhlet extraction apparatus.

SAMPLE PREPARATION

41

Solvent evaporates from the round bottomed flask, rises through the extractor, condenses in the condenser, drips back into the extractor, and gradually immerses the contents of the thimble. As the level of solvent reaches a critical volume in the extractor, the apparatus design is such that all of the solvent in the extractor drains back into the round bottomed flask leaving behind solid material in the thimble. This

Figure 2.14 Solvent Assisted Flavor Evaporation (SAFE) apparatus developed by Engel et al. [32] improved upon pre-existing simultaneous distillation and solvent extraction techniques.

42


process is repeated many times over several hours to provide multiple extractions of the solid material present in the thimble. Ideally suited to remove lipid material from a solid, depending on the solvent used this method may also be used to extract other components.

2.8.3 Solvent Assisted Flavor Evaporation (SAFE) Engel et al. [32] described Solvent Assisted Flavor Evaporation (Figure 2.14), a method that combined vacuum distillation, cold trapping, and in some cases solvent extraction (solid samples such as popcorn, coffee or bread crust required solvent extraction). This method has proven useful and a step above previous simultaneous distillation extraction (SDE) procedures. What distinguishes this method is it is faster, safer, and easier to maintain temperature controls than previous methods that employed similar high vacuum, cold trapping conditions. Additionally it yields an extract that minimizes formation of off-flavor artifacts and provides a close approximation to authentic flavor. In addition to Engel’s original work, another good review of this technique is found in Werkoff et al. [33].

2.9 SUMMARY Due to a greater requirement for rapid sample screening, sample prep techniques for subsequent GC separation have necessarily become automated to allow multisample analysis. Tools that lend themselves to automation such as SPME, Twister, and dynamic headspace extraction traps will be utilized more frequently in the coming years as demand for basic chemical information continues to increase. The novice scientist must be able to utilize such techniques in order to be an asset in industry and academia. The informed scientist must also understand fundamental extraction methodology because principles of future techniques that meet the rapid and automated criteria will be based on tried and true methods including static, dynamic, solvent, and distillation extractions.

REFERENCES 1. Micro Total Analysis Systems, Silly Putty, and Fluorous Peptides http://www.fluorous .com/journal/?p=86. 2. Lee, J. N., Park, C., and Whitesides, G. M. (2003) Solvent compatibility of poly(dimethylsiloxane)-based microfluidic devices. Anal. Chem., 75, 6544.

SAMPLE PREPARATION

43

3. Lotters, J. C., Olthuis, W., Veltink, P. H., and Bergveld, P. (1997) The mechanical properties of the rubber elastic polymer polydimethylsiloxane for sensor applications. J. Micromech. Microeng., 7, 145–147. 4. David, E.T., Tienpont, B., Sandra, P. (2003) Stir-bar sorptive extraction of trace organic compounds from aqueous matrices. LC/GC, 21, 109. 5. Reineccius, G. (2006) Flavor Chemistry and Technology. CRC Press, Boca Raton, FL, USA, pp. 33–72. 6. Arthur, C. L., and Pawliszyn, J. (1990) Solid phase microextraction with thermal desorption using fused silica optical fibers. Anal.Chem., 62, 2145. 7. Marsili, R.T. (1999) SPME-MS-MVA as an electronic nose for the study of off-flavors in milk. J. Agric. Food Chem., 47, 648– 654. 8. Marsili, R. (1999) Comparison of solid phase microextraction and dynamic headspace methods for the gas chromatographic – mass spectrometric analysis of light induced lipid oxidation products in milk. J.Chrom. Sci., 37, 17–23. 9. Rouseff, R. L. and Cadwallader, K. R. (2001) Headspace Analysis of Foods and Flavors: Theory and Practice. Kleur Publishing, New York, USA, pp. 212. 10. Rouseff, R., Bazemore, R.; Goodner, K., and Naim, M. (2000) GC-Olfactometry with Solid Phase Microextraction of Aroma Volatiles from Heated and Unheated Orange Juice. In Headspace Analysis of Foods and Flavors: Theory and Practice (eds R. L. Rouseff and K. R. Cadwallader), Plenum, New York, USA. 11. Koziel, J., Cai, L., Wright, D., and Hoff, S. (2007) Solid phase microextraction as a novel air sampling technology for improved, GC-olfactometry based, assessment of livestock odors. In press J. Chrom. Sci. 12. Wright, D. et al. (2005) Multidimensional gas chromatography-olfactometry for the identification and prioritization of malodors from confined animal feeding operations J. Agric. Food Chem., 53, 8663– 8672. 13. Wright, D.W. et al. (2004) Multidimensional gas chromatography-olfactometry based investigations of odor quality issues in packaging and consumer products. Proceedings of TAPPI Conference, Indianapolis, Indiana, USA. TAPPI. 14. Shirey, R. (1999) Sample prep made easy. SPME Newsletter Winter 1999/2000. 15. Anonymous (1998) Solid phase microextraction: theory and optimization of conditions. Supleco Bulletin, 923, 1–8. 16. Zhang, Z. and Pawliszyn, J. (1993) Headspace solid phase microextraction. Anal.Chem., 65, 1843–52. 17. Payne, R., Labows, J., and Liu, X. (2000) Released oral malodors measured by solid phase microextraction/gas chromatography mass spectrometry (HS-SPME-GC-MS). Proceedings of ACS Flavor Release No. 0841236925. American Chemical Society, Washington, DC, USA. 18. Baltussen, E., Sandra, P., David, F., and Cramers, C. (1999) Stir bar sorptive extraction (SBSE), a novel extraction technique for aqueous samples: Theory and principles. J. Microcol., 11, 737– 747. 19. Soini, H. A. et al. (2005) Stir bar sorptive extraction: a new quantitative and comprehensive sampling technique for determination of chemical signal profiles from biological media J. Chem. Ecol., 31, 377– 392. 20. Soini, H. et al. (2005) Comparative investigation of the volatile urinary profiles in different Phodopus hamster species. J. Chem. Ecol., 31, 1125– 1143. ˜ R., and Sandra, P. (2003) Analysis of volatiles 21. Demyttenaere, J. C. R., Morina, of malt whisky by solid-phase microextraction and stir bar sorptive extraction. J. Chrom. A., 985, 221–232.

44


22. Kishimoto, T., Wanikawa, A., Kagami, N., and Kawatsura, K. (2005) Analysis of hop-derived terpenoids in beer and evaluation of their behavior using the stir bar-sorptive extraction method with GC-MS. Agric. Food Chem., 53, 4701– 4707. ´ 23. Leon, V. M. et al. (2003) Analysis of 35 priority semivolatile compounds in water by stir bar sorptive extraction-thermal desorption-gas chromatography-mass spectrometry I. Method optimization. J. Chrom. A., 999, 91–101. 24. Gurbuz, O. and Rouseff, R. (2005) Analysis of aroma active compounds in wine using stir bar sorptive extraction with GC-olfactometry. Proceedings of the 56th Citrus Processor’s Meeting, October 20, 2005, Lake Alfred, FL, USA. 25. David, F. and Sandra, P. (2007). Stir bar sorptive extraction for trace analysis. J. Chrom. A., 1152, 54–69. 26. Loughrin, J. (2006) Comparison of solid-phase microextraction and stir bar sorptive extraction for the quantification of malodors in wastewater. J. Agric. Food Chem., 54, 3237– 3241. 27. Bicchi, C., Iori, C., Rubiolo, P., and Sandra, P. (2002) Headspace sorptive extraction (HSSE), stir bar sorptive extraction (SBSE), and solid phase microextraction (SPME) applied to the analysis of roasted arabica coffee and coffee brew. J. Agric. Food Chem., 50, 449– 459. 28. Marsili, R. and Laskonis, L. (2006) Evaluation of PDMS-based extractiontechniques and GC-TOFMS for the analysis of off-flavor chemicals in beer. Gerstel application note 10/2006. 29. Pfannkoch, E. and Whitecavage, J. (2006) Elimination of non-volatile sample matrix components after GC injection using a thermal desorber and microvial inserts. Gerstel application note 4/2006. 30. Parliament, T. H. (1986) A new technique for GLC sample preparation using a novel extraction device. Perfum. Flavor., 11, 1–8. 31. Jella, P., Rouseff, R., Goodner, K., and Widmer, W. (1998). Determination of key flavor components in methylene chloride extracts from processed grapefruit juice. J. Agric. Food Chem., 46, 242– 247. 32. Engel, W., Bahr, W., and Schieberle, P. (1999) Solvent assisted flavor evaporation – a new and versatile technique for the careful and direct isolation of aroma compounds from complex food matrices. Eur. Food Res. Technol,, 209, 237– 241. 33. Werkhoff, P., Brennecke, S., Bretschneider, W., and Bertram, H.J. (2002) In Flavor: Fragrance, and Odor Analysis (ed. R. Marsili), Marcel Dekker, New York, USA, p. 139.

3 Traditional Flavor and Fragrance Analysis of Raw Materials and Finished Products Russell Rouseff IFAS, Citrus Research and Education Center, University of Florida, Lake Alfred, USA

Kevin Goodner Sensus, LLC, Hamilton, USA

3.1 OVERVIEW Most flavor samples have a biological origin and thus typically consist of complex chemical mixtures with a wide range of chemical and physical properties which require separation. In addition, flavor active components are typically trace components which require that the compounds of interest need to be extracted and concentrated in order to simplify the analysis and to raise the concentration of the components of interest above instrumental detection limits. Specific details of this Practical Analysis of Flavor and Fragrance Materials, First Edition. Edited by Kevin Goodner and Russell Rouseff. © 2011 Blackwell Publishing Ltd. Published 2011 by Blackwell Publishing Ltd.

46


critical step have been presented in the proceeding chapter. However, in discussing analytical procedures, it cannot be overemphasized that the most sophisticated analytical procedures cannot compensate for an inadequate extraction. The objective of the analysis process is to separate an extract into its constituent chemicals to determine either their nature (qualitative analysis) or their amounts (quantitative analysis). In the flavor industry the acquired data is used by various groups within a company to satisfy specific needs. For example, purchasing decisions are predicated on a material’s ability to meet certain analytical specifications that may include standards of identity, authenticity, chemical purity, and overall component profile. The manufacturing sector relies upon analytical information to control processes and assess equipment effectiveness. Research groups require the ability to identify and quantify unique chemicals in order to measure the completeness of an anticipated reaction. Quality control groups would be most interested in determining the major flavor impact compounds in order to maintain a consistent flavor profile as well as identifying off-flavors that occasionally appear. Figure 3.1 illustrates how analytical chemistry associates with nearly every facet of a flavor and fragrance company. The abundance, complexity, and application of analytical methodology have significantly increased with the advancement of instrumental technologies with the associated computer hardware and software. Almost all modern instruments employ computers to control the instrumental processes and record data thus allowing for unattended analysis and automation. Modern analytical laboratories are typically equipped with a wide array of instrumentation capable of trace analysis.

Research

Product optimization

Technical support Quality control Development

Compounding ANALYTICAL Duplication

In-Process control

Purchasing

Figure 3.1 Reliance on analytical support throughout the various departments of a flavor and fragrance company.

TRADITIONAL FLAVOR AND FRAGRANCE ANALYSIS

47

One of the major goals for flavor and fragrance companies is to produce high quality products with consistent sensory properties. Analytical tools have altered the once exclusively intuitive flavor/fragrance creation processes towards one of more calculated science based upon chemical composition. The increased ability to determine a product’s complete composition analytically has motivated flavorists and perfumers to increase the sophistication of their formulations to achieve differentiation and thwart duplication from the competition. This is typically achieved by adding potent, high sensory-impact components that are difficult to quantify and identify chemically. At one time sensory evaluation was the exclusive means of assessing the quality and consistency of a given flavor or fragrance and it still is a major assessment tool. Today, sensory analysis has been augmented with detailed analytical information which is particularly useful when determining subtle differences in sensory profiles. It was the only available tool prior to the development of analytical instrumentation. Sensory analysis still remains the final checkpoint for a developer or quality control analyst; however, analytical equipment provides objective, repeatable, and reliable assessment. Traditional flavor and fragrance analysis involves the evaluation of physical, chemical, and sensory characteristics. This chapter covers the analytical aspects, so sensory techniques will be discussed later in the book. Prior to gas and liquid chromatography, classical methods focused on basic physical attributes such as color and clarity. More objective methods evolved to include specific gravity, optical rotation, refractive index, and so on. These tools were employed to determine the authenticity of essential oils and other flavoring materials. Today, methods are specifically tailored to identify and quantify individual chemicals within every imaginable matrix (i.e., living material, extract, headspace, finished flavor and fragrances, store-bought beverage, etc.). This shift from physical to chemical attribute evaluation has been made possible because of greater instrumental selectivity for specific components, improved detector sensitivity, and computer controlled instrumental automation.

3.2 PHYSICAL ATTRIBUTE EVALUATION There are a number of physical characteristics of products that are important to the food and flavor industry. Generally, product appearance, moisture content, and concentration are some of the most important and most often measured.

48


3.2.1 Color – Optical Methods One of the primary methods consumers use to evaluate food and beverages is their appearance. Therefore, the color of the final product – and, by inference, its ingredients – is a critical attribute to determine. There are two generally used measurements for determining color: absorbance of light at a specific wavelength or by measuring the color using the LAB system (discussed below). Specific wavelengths are sometimes used to describe the intensity of color of a product. For example, caramel color manufacturers Sethness Products and D.D.W. Williamson, two leading caramel color manufactures, each use the absorbance of a 0.1 % weight/volume solution but at differing wavelengths. Sethness Products describes their products based on the absorbance at 560 nm while D.D.W. Williamson uses 610 nm. Both companies use the hue index to describe the redness of their products. The hue index is 10 times the logarithm of the absorbance at 510 nm divided by the absorbance at 610 nm [1]. A510 Hue = 101Log A610 These two values – intensity and hue – are the main appearance characteristics used in the caramel color industry. Another example is in the brewing industry. One of the main methods is the Standard Reference Method, which is defined by SRM = 12.7 ∗ D ∗ A430 where D is a dilution factor. This method yields numbers generally between 2 and 70 which would correspond to a light yellow colored beverage to a very dark brew. These two industries are just examples of some of the methods that specific products use involving absorption. While some companies and researchers use specific wavelengths, others use the LAB color space. The Hunter 1948 LAB color space is designed to approximate human vision using the variables L, a, and b. However, the Hunter 1948 method has largely been replaced with the CIE 1976 L∗ , a∗ , b∗ space where CIE is the abbreviation for the International Commission on Illumination [2]. In general, the L∗ value matches the human perception of lightness (L∗ = 0 is black and L∗ = 100 is white), a∗ is a measure of the red/green of a sample (i.e., negative a∗ values mean green and positive values red), and b∗ is a measure of the yellow/blue


49

of a sample (i.e., negative b∗ values mean blue and positive values mean yellow). A common colorimeter used by academics and the food and flavor industry is the Minolta CR-400 series colorimeter. This instrument is a portable, handheld unit that can determine L∗ , a∗ , b∗ values of products by using reflectance or transmittance with an adapter. Given the history of the LAB system, it is no surprise that Hunter Lab systems are also popular colorimeters. These systems and others can provide analytical data on the color of ingredients or final products. One thing to keep in mind, whether using a spectrometer to measure specific wavelengths or a colorimeter to obtain L∗ , a∗ , and b∗ values, these numbers are useful for quality control/quality assurance, research and development, and product development, but do not indicate consumer acceptance. Sensory testing (Chapter 8) is critical to determine consumer acceptance.

3.2.2 Turbidity While color is an obviously critical component in food products, another important factor in beverages is turbidity (i.e., clarity) [3]. The desired clarity of a product varies depending on the product and even varies within a product category. For example, some fruit juices are clarified and the consumer has come to expect that (e.g., apple and grape) while others are not (e.g., orange juice). Other examples include tea and coffee, where different brands have different turbidity levels ranging from clear to highly turbid. Turbidity is measured by how much a sample will scatter light passing through it. Measuring turbidity in this manner uses a nephelometer which reports nephelometric turbidity units (NTU). Turbidity tests are generally performed on relatively inexpensive equipment and only take a couple of minutes to obtain results.

3.2.3 Water Activity Water activity (Aw ) is an important physical characteristic of a product. Water activity (Aw ) is a measure of the energy of the water in a sample and it has no units. It is often described as measuring the ‘available’ water in a system. This description is used to explain the situation where two products have the same water content, but different water activity, or different moisture content and the same water activity. For example, pasta with a moisture content of 12 % could have a Aw of

50


0.50 while rolled oats with 10 % moisture could have a Aw of 0.70. One of the reasons water activity is so important in the food industry is that microorganisms will only grow if the water activity is acceptable. For example, many microorganisms cannot grow if the water activity is below 0.9, most molds require a water activity above 0.8, and a water activity below 0.6 will inhibit all microbial growth [4]. There are two main methods for determining water activity, the chilled mirror method and the capacitance method. The chilled mirror method uses a mirror which is chilled until dew is formed and detected with an optical sensor. The capacitance instruments use two charged plates with a polymer membrane between them. Instruments vary in price but generally start around $2000 and can determine the water activity of a sample in a few minutes.

3.2.4 Moisture Content The moisture content of samples is another commonly measured parameter. There are several ways to determine this, but loss on drying (or evaporative residue) is probably the most common. This can be done in an oven or using bench top instrumentation. Generally the oven is cheaper and can dry more samples at once, but it takes longer to get results and is more manual, meaning a technician has to weight the sample before and after drying and perform the calculations. Additionally, oven drying can be problematic for some samples if they decompose or char because if there is only one oven but differing conditions are needed for various samples. These problems can be ameliorated by using a bench top automated moisture analyzer. Typical instruments today can have multiple methods for different types of samples and each method can be optimized to minimize drying time. Additionally, some units will use preliminary data to extrapolate to an end point, further decreasing analysis time and preventing charring or decomposition. Bench top moisture analyzers cost somewhere between $2000 and $20 000 depending on features such as balance accuracy/precision or using microwave energy to speed the heating. 3.2.4.1 Karl Fischer Method While loss on drying is the most common and one of the most accepted methods for determining moisture, there are other methods. The only other primary moisture determination method is the Karl Fischer titration. The Karl Fischer titration is named after the German chemist Karl


51

Fischer who developed the method in 1935 [5]. The advantage of the Karl Fischer method is that it is specific for water while loss on drying will measure the loss of any volatile compounds (e.g., flavor volatiles). The Karl Fischer method is also used for very low moisture content samples as it is the most precise, accurate, and sensitive method. The titration uses an alcohol, base, sulfur dioxide, and iodine. The reactions that occur in the sample are listed below. CH3 OH + SO2 + C3 H4 N2 → (C3 H4 N2 · H) · SO3 CH3 (C3 H4 N2 · H) · SO3 CH3 + 2 · C3 H4 N2 + I2 + H2 O → (C3 H4 N2 · H) · SO4 CH3 + 2(C3 H4 N2 · H)I In current laboratories the Karl Fischer method is most commonly performed with an automated system which is available from several manufacturers for approximately $10 000. 3.2.4.2 Secondary Moisture Determination Methods While loss on drying and the Karl Fischer titration are considered primary moisture analyzers, there are several so-called secondary moisture determination methods. One method is to measure the water activity and determine the moisture content. This method is considered secondary as it can only be used once a relationship has been generated between water activity and moisture content, as determined by a primary method, and water activity. This relationship between moisture content and water activity has to be determined for every product. A new product would have to have the relationship between water activity and moisture determined. While that is a negative, such instruments determine water activity and moisture content in a single analysis. Another secondary method is by utilizing the near infrared (NIR) spectrum. Much like using water activity to measure moisture content, a calibration curve has to be generated for each product using the NIR spectra and moisture content from a primary moisture analysis (loss on drying or Karl Fischer). While the instrumentation is generally considerably more expensive than the other methods, results are available in seconds.

3.2.5 Optical Rotation Optical rotation is a parameter which is commonly used in the flavor industry as a measure of identity. If plane polarized light is passed

52


through solutions, it can become rotated either clockwise or counterclockwise [6]. Measuring this optical rotation can help determine the concentration and identity of a compound in solution. For this reason, it is a fairly common analysis of flavors to ensure that the concentration and identity are within the specifications provided by the manufacturer. Instruments to measure the optical rotation are generally available for around $10 000 for a bench top model.

3.2.6 Specific Gravity Specific gravity, or relative density, is the ratio of the density of an item to that of a reference material. Most often, the reference material is water. The temperature at which the density measurement is determined must be specified for both materials. A common temperature for the reference material of water is 4 ◦ C as the density of water at that temperature is 1.00 g/mL and therefore the density of the sample is equal to the specific gravity. In this manner, many people treat specific gravity to be the same as density (due to the convenient choosing of the reference). Density and/or specific gravity are both properties that are used to determine the purity of a product. There are many ways to determine specific gravity and/or density. The simplest and probably least accurate is to use a graduated cylinder to measure a volume of material which divided into the weight will provide the density of liquid. A more accurate method, but not without issues, particularly for viscous fluids is to use a volumetric flask in the same matter as the graduated cylinder. An item of specialized glassware called a pycnometer is a small vessel with a lid that has a small hole in it such that the vessel can be filled and excess material is eliminated through the small hole. The weights of the sample and water are then divided to determine the specific gravity. There is also instrumentation that can determine density and specific gravity. There are handheld units that cost a few thousand dollars and bench top models which are more precise and accurate but cost considerably more.

3.2.7 Refractive Index Refractive index is a measure of how much the speed of light is reduced in an item. As light passes from one medium to another, the light will change direction. The greater the difference in velocity of light through the two media, the greater the observed angle [7]. Refractive index


53

is measured using a refractometer which can vary between handheld and bench top using either transmittance or reflectance measurement. As refractive index is an inherent property of a substance, it is commonly used to confirm the purity of samples, particularly in a QA/QC (quality assurance/quality control) type of environment. One particularly common use of refractive index is to measure the concentration of aqueous solutions. The refractive index of a solution will change as the concentration of the solute changes. This relationship can be measured and used to determine the concentration from the refractive index. Modern refractometers will automatically apply the relationship and displace the concentration of many analytes. Some examples include: ethanol content, high fructose corn syrup (HFCS), salinity, and most commonly, ◦ Brix (sucrose concentration). This last relationship is used by many in the industry for many types of products – some of them have no sucrose at all. For example, many people measure tea and coffee solids as ◦ Brix. In a sucrose solution, a 50 ◦ Brix solution is 50 % solids and 50 % water. For other materials, such as tea and coffee, that is not the case. For example, a 50 ◦ Brix tea or coffee concentrate will have approximately 40 % solids and 60 % water. However, while the measurement is not accurate, it is consistent and has been used for many years in the tea and coffee industry and therefore is not likely to be replaced.

3.2.8 Sugars/Soluble Solids When dealing with sugar solutions, ◦ B (degrees Brix) is used to describe the amount of dissolved sucrose. This is common in fruit juices, wines, and some other products [8]. ◦ Brix is generally determined by measuring a physical property and relating that to the percentage of sucrose in a solution. This physical property can be various properties such as density, specific gravity, refractive index, or infrared vibrational wavelengths. This means that any solution can be analyzed to determine a ◦ B measurement, even if little to no sucrose is in solution. While this might at first seem to be a totally erroneous method, it is not necessarily so. Consider that as the concentration of a solute (or in the case of complex mixture, solutes) increase, the refractive index will change. This will provide an apparent ◦ B reading. This ◦ B reading can either be used directly or related to actual concentration of solutes (which can be determined by drying or percent moisture analyses). However, using refractive index to determine ◦ B is predicated upon one important assumption – that all solutes are actually in solution (and

54


thus affect refractive index) and not suspended (and thus do not affect refractive index). Additionally, there is the danger of suspended solids settling on the instrument’s prism and interfering with the measurement. Density and specific gravity might solve the problems of suspended solids settling on the prism and possibly of suspended solids being correctly accounted for. There is the very real possibility that the density would be different if the solutes are suspended as opposed to being in solution. However, for many compounds the effects are very small and likely within experimental error. Compounds that dramatically affect density do so generally by affecting the volume of the solution. While this does occur it is usually the exception and not the rule. Infrared is specific for sugar vibration bands and therefore would not work for other solutes.

3.2.9 Viscosity While many physical properties are useful for describing a product, possibly one of the more important for processing is viscosity. Many pumps and other equipment will not work if the viscosity is too high. Viscosity is a measure of a product’s fluidity [9]. The lower the viscosity, the easier the product will flow. Viscosity can be measured with a viscometer which can range from glassware to bench top units. Viscosity is often reported in units of centipoise. However, this explanation of viscosity is really only skimming the surface of this vast topic. Whole books have been written solely on viscosity and one should explore those if viscosity is important to their product.

3.3 INSTRUMENTAL ANALYSIS There are a wide range of instrumental techniques available to today’s analytical flavor chemist. The technique employed will depend primarily on what information is desired and what techniques are available. Pure compounds will be treated differently than complex mixtures. In addition the physical state, volatility, solubility, and sample size are additional features which may exclude some techniques. Gas and liquid chromatography will be the major separation techniques discussed in this chapter. Major identification techniques will include mass spectrometry and nuclear magnetic resonance (NMR) as they are the most commonly employed techniques.


55

3.3.1 Separation Techniques 3.3.1.1 Gas Chromatography (GC) Although there are many types of chromatography, high resolution capillary gas chromatography is by far the major technique employed to separate flavor materials which can be volatilized. It is not uncommon to have hundreds of components in a wide range of concentrations in a flavor sample. Good separation is difficult but required to obtain an accurate identification. Coeluting substances add extraneous data which will reduce identification confidence or in extreme cases result in a false identification. 3.3.1.2 GC Retention Data The time it takes for a specific volatile to elute from the end of a GC column after injection is a characteristic of the volatile and its interaction with the column stationary phase. This time is also a function of several GC parameters such as carrier gas flow rate, column length, and oven temperature. It is called either elution or retention time, with the latter term being more widely used. Although this time is characteristic of each volatile, retention time is not a unique measurement and many compounds will potentially share the same retention time. For this reason, it should be kept in mind that GC is primarily a separation technique and not an identification technique. Although chromatographic retention time has been used as an identification technique in earlier literature, it is no longer considered acceptable practice. To reduce the number of potential volatiles that will have similar retention times, a second retention time is determined with a column stationary phase different from the first. If retention times for both a standard and unknown match from two distinctly different column stationary phases, then it is highly likely they are identical, but not an absolute identification. 3.3.1.3 Standardized Retention Index Systems Since there are so many variables in GC, an absolute time measurement is of limited value as column conditions tend to drift with time and use and they are therefore difficult to duplicate. To allow for a more uniform system of reporting retention behavior, systems based on relative retention of the compound of interest compared to a series of standards

56


have been developed. In this way data from columns of different lengths and carrier gas flow rates could be compared. However, it needs to be emphasized that each system is valid only for the stationary phase from which the data is obtained and cannot be directly used for columns with different stationary phases. One of the first to be developed was ´ in the late 1950s and was based on one popularized by Ervin Kovats a homologous series of linear alkanes [10]. This system ran a series of linear alkanes from C5–C25 where each alkane was given a value equal to the number of carbon atoms times 100. Each new volatile was given a value by interpolating between the two nearest alkane values. However, this system is based on isothermal temperature systems which are rarely employed today because of the wide range of volatilities in most flavor samples. This approach has been modified to allow for the use of linear temperature gradients [11]. The term ‘Kovats Index’ has been largely supplanted by the term ‘retention index’ or ‘linear retention index’ (LRI) for values obtained under thermal gradient conditions. An example is shown in Figure 3.2 in the case of octanol separated on a DB-5 column. In this situation octanol elutes between decane (C10) and undecance (C11). Its LRI value can be calculated using the retention times from the chromatographic report as shown in the general equation: Rt(peak of interest) − Rt(preceding alkane) LRI = 100 Rt(following alkane) − Rt(preceding alkane) + #C(preceeding alkane) In the case of octanol shown in Figure 3.2: 9.22 − 7.77 LRI = 100 + 10 = 1071 9.82 − 7.77 3.3.1.4 GC Injection GC injections are generally of either solvent or solventless types. Earlier chromatographic studies primarily employed solvent based liquid injections as solvents were typically used to extract the compounds of interest from the matrix and/or then concentrated by carefully removing the solvent to concentrate the volatiles of interest. Injections of liquid extracts have the advantage of knowing exactly how much material is put onto the column (if in the preferred splitless mode). In the case of standards, both internal and external standardization procedures can be

57

C20

C19

C18

C17

C16

C15

C14

C13

Octanol

FID Response

C10

C11

C12


8

6

10

12

14

16

18

20

22

24

Time – min (DB-5)

4

6

8

10 12 14 Time – min (WAX)

C20

C19

C18

C17

Octanol C16

C15

C14

C13

C12

C11

FID Response

C10

Figure 3.2(a) Mixture of linear alkane standards and standard octanol separated on a 30 m 5 % phenyl methylsiloxane column, where C10 = decane, C11 = undecane, C12 = dodecane, and so on.

16

18

20

Figure 3.2(b) Homologous series of alkane standards with octanol, separated on a 30 m wax column.

used. Liquid injection is also simpler, more dependable, and less expensive to automate than solventless techniques. The major disadvantage of liquid injection is that it often introduces material onto the column that may be of limited volatility and over many injections will degrade

58


column performance. One way to minimize this problem is to clean up the extract using either distillation/condensation or classic column chromatography prior to concentration and injection. In addition, the solvent chosen must have a boiling point at least 20 ◦ C lower than the most volatile component of interest. The most commonly employed solvents are either pentane or ethyl ether or a mixture of the two as they have relatively low boiling points and are still liquids at room temperature. A major disadvantage to solvent based injection is the large solvent peak which coelutes with those volatiles that are only slightly retained by the column; rendering analysis of these compounds virtually impossible. Solventless techniques include fixed volume headspace injection, thermal desorption purge and trap and solid phase micro extraction (SPME) developed by Janusz Pawliszyn [12]. Of these SPME has become increasingly popular as standard GC injectors can be employed and the injection apparatus is simple and relatively inexpensive compared to the alternatives. SPME has been well described in the previous chapter on sample preparation and additional details can be found there. All solventless injection systems allow for the detection of highly volatile components normally blocked out by the solvent peak. The disadvantage of SPME is that the fibers are fragile and easily broken. In addition, quantification can be difficult as the amount sorbed on the fiber is time, temperature, and matrix dependent. Adding known amounts of standards to the matrix is the recommended way to quantify specific volatiles in samples with complex matrices. 3.3.1.5 GC Columns (Stationary Phases) There are a wide range of column types available for capillary gas chromatographic separations. A few of the most common are listed in Table 3.1. The 5 % phenyl methylpolysiloxane columns are probably the most widely employed phase. More standard retention index values are listed for this stationary phase than any other. It is a rugged column that has slightly better separation power than the previously popular 100 % methylpolysiloxane. Another reason it is widely used is that it produces reasonably reproducible retention index values so that researchers can compare their data with that found in the literature with a fair degree of confidence. Wax (polyethylene glycol)/Free Fatty Acid columns produce the best separations for oxygenated (more polar) volatiles. Flavor chemists are typically more interested in oxygenated volatiles than hydrocarbon

59


Table 3.1 Comparison of common stationary phases used in capillary gas chromatography. Stationary phase

Temperature range (◦ C)

Methylpolysiloxane

50–325

5% Phenyl methylpolysiloxane

50–325

50% Phenyl methylpolysiloxane

40–325

Polyethylene glycol

20–260

Free fatty acid

20–260

Porous silica or modified alumina or porous DVB∗

−80–300

Recommended uses

Trade names

Nonpolar compounds – low selectivity, volatiles separated by increasing boiling point. Good thermal stability Most widely used phase. Best for nonpolars, similar to methyl polysiloxane, but more selectivity due to phenyl content. Rugged column, good thermal stability Best for high boiling flavones, coumarins, and steroids. Additional selectivity due to higher phenyl content. Excellent thermal stability Best for polar compounds containing oxygen. Susceptible to oxygen degradation.

DB-1, SE-30, OV-1, OV-101, HP-1, RTx-1

DB-5, SE-54, OV-23, HP-5, RTx-5,

DB-17, OV-17, HP-17, RTx-17,

Carbowax 20M, DB-Wax, Stabilwax, BP-20, HP-20M, AT-Wax Especially useful for OV-351, HP-FFAP, volatile fatty acids and SP-1000, fatty acid methyl esters AT-1000, (FAMEs). Good thermal stability but susceptible to oxygen degradation Best for gases and low Gas Pro, molecular weight GC-PLOT, hydrocarbons HP-PLOT

Note: DVB∗ = divinyl benzene polymer.

terpenes as the oxygenated species are generally more aroma active. Unfortunately, the retention index values from these column types tend to be much more dependent on coating thickness, oven gradient rates and injection history than those columns which employ methyl polysiloxane

60


such as DB-1 or DB-5. Therefore, although these columns are commonly employed, their retention index values cannot be as readily compared with those that can be found in the literature. Another column type of interest to flavor analysts when it might be necessary to separate highly volatile aroma active volatiles like hydrogen sulfide or methanethiol is the porous layer open tubular column typically called a PLOT column. As indicated in Table 3.1, the material employed for this type of column can vary widely. PLOT columns are columns with thick interior wall coatings of porous material designed to retain highly volatile materials not normally retained by the major column types. The coating consists of particulate materials which offer a range of unique selectivities for volatiles which are gases at room temperature. They are, however, highly retentive and should not be routinely used for separating samples which contain compounds of low volatility. Even though these columns have a high temperature range to allow elution of highly retained material, the chance of contamination or increased background levels due to previous injections is fairly high. These columns are best used for headspace analyses. An example PLOT separation is shown in Figure 3.3 for a refinery gas sample. Few if any refinery gases are aroma active, but the chromatogram illustrates the ability of the PLOT column to separate highly volatile components. The PLOT column can separate highly volatile hydrogen sulfide, methane thiol and carbon disulfide. 3.3.1.6 GC Detectors Two basic detector types are commonly employed in chromatographic systems, general mass detectors which respond to most or all volatiles and selective detectors which are specific for a desired subset of volatiles. General mass detectors include the common FID detector and the total ion current (TIC), MS detector (MS-scan) as listed in Table 3.2. General mass detectors are necessary to determine column loading and to help quantify the volatiles present in highest concentrations. Selective detectors are employed when complex samples are analyzed and coelution is a major problem. In these cases the detector should respond only to the compounds of interest and it would not be necessary to chromatographically separate all interferences from the compounds of interest unless the interfering volatiles are in sufficiently high concentration to prevent the selective detector from operating properly. Examples would include fluorescence quenching for optical systems such as the PFPD or

61

TRADITIONAL FLAVOR AND FRAGRANCE ANALYSIS 1. Methane 2. Ethane 3. Ethylene 4. Propane 5. Propylane 6. Isobutane 7. n-butane 8. Propadiene 9. Acetylene 10. Trans-2-butene 11. 1-butene 12. Isobutylene 13. Cis-2-butene 14. Isopentane 15. n -pentane 16. 1,3-butadiene 17. Trans-2-pentene 18. 2-methyl-2-butene 19. 1-pentene 20. Cis-2-pentene 21. Hexanes

6

4

12

7 3 5 10 11 13

16

12 14 15 8

0

2

4

6

9

17 19 18 20 21 8

10

12

14 min

Figure 3.3 Separation of refinery gases using a PLOT column. Column = 50 m, 0.53 mm id, 10 μm Rt®-Alumina BOND/Na2 SO4 , hydrogen carrier gas at 8.0 psi. FID detection. Oven temperature program = 45 ◦ C (hold for 1 min) then to 200 ◦ C at 10 ◦ C/min. Courtesy: Restek.

ion–molecule interactions with MS in the SIM mode. Selective detectors often have greater sensitivity (lower detection limits) than many general mass detectors. The detectors listed in Table 3.2 are far from inclusive but contain the detectors most commonly used in flavor analysis. Even though these detectors are ranked in terms of diminishing linear range, the sensitivity (minimum detection limits) is equally or more important. The human nose is a highly sensitive detector whose detection limits for some volatiles are difficult to match with instruments. In some cases the human nose is more sensitive than the best analytical detectors. Examples include the sulfur containing methional (cooked potato) and the chlorinated phenol, 2,4,6-Trichlorophenol, TCA (musty) among others. Shown in Figure 3.4 is the comparison of a sulfur specific detector to that of a more general carbon PFPD detector. Three of the more important coffee volatiles are separated and quantified using the sulfur specific detector including the coffee character impact compound 2-furfuryl thiol (A). The upper trace of this chromatogram is specific for sulfur volatiles. The lower trace is a more general detection of carbon volatiles.

62


Table 3.2 Comparison and characteristics of common GC detectors sorted by linear dynamic range. Type

Selective for:

FID

General mass

NPD

Selective

Volatiles that ionize in air-H2 flame N, P

PFPD MS-scan MS-SIM ECD FPD

Selective General mass Selective Selective Selective

Primarily P,S Most organic volatiles Specific m/z ions Primarily halogens Primarily P,S

AES

Selective

Most elements

PFPD S

Name

Detection limits

Linear range

5 pg C/s

>107

0.4 pg N/s 0.2 pg P/s 0.1 pg S/s 10 ng 10 pg ∼0.1 pg Cl/s 20 pg S/s 0.9 pg P/s 0.1–1 ng element specific

106 105 105 105 104 103 103

A

C

PFPD C

B

16

18

20

22

24

26

28

30

Time – (min-wax)

Figure 3.4 Aroma volatiles from ground coffee comparing the response from a sulfur specific detector to that of a carbon specific detector: A = 2-furfuryl thiol, B = methional, C = furfuryl methylsulfide 30 m × 0.32 mm id wax column, oven program = 60–180 ◦ C at 3 ◦ /min and with a 5 min hold at 180 ◦ C.


63

3.3.2 Identification Techniques 3.3.2.1 Retention Index Approach The time it takes for each volatile to pass through columns with various chromatographic stationary phases is a characteristic of that substance and the stationary phase. Unfortunately it is not a unique characteristic as many compounds can share very similar or identical retention values. Determining this time as a retention index value does not reduce the number of compounds that share this same value. The general rule of thumb for identification is to have two independent tests confirm the identity of an unknown. Therefore a single retention index match cannot be used for identification. The second piece of independent information would be matching the unknown with a standard on another column of very different chromatographic properties. Typically, the match is made with something like a DB-5 column and a wax column as shown in Figures 3.2(a) and (b) and for the case of octanol. In Figure 3.2(a), octanol elutes between decane (C10) and undecane (C11) on the nonpolar DB-5 column with an LRI value of 1071. When the somewhat polar octanol is chromatographed on a polar wax column, it is more strongly retained and elutes between C15 and C16 with an LRI value of 1565. If the unknown of interest also had an LRI value of something like 1070, it would indicate that the unknown peak could be octanol, but certainly does not prove it, as many other compounds have similar values. However, if the unknown also had a wax LRI of 1565, then it was quite likely to be octanol as few if any other compounds would possess similar values on both dissimilar chromatographic columns. It is always recommended to match retention index values of unknowns with standards run under identical conditions. However, it is impossible for most flavor analytical chemists to have at their disposal, much less test, the thousands of volatiles known to exist. This is where standard tables of compounds with their LRI values become extremely useful as they can narrow the potential list of standards to be tested. It should be pointed out that the values found in these tables are similar, but rarely agree perfectly. For example, the sample octanol had a value of 1071 on a DB-5 column. This can be compared with values of 1066 and 1078 from the Reading University site, a value of 1070 from the Pherobase site and 1068 for the University of Florida site. Again the DB-5 column value is chosen to make an initial tentative identification as LRI variability for it is much less than that of wax. Standard octanol

64


Table 3.3 LRI data bases. Books Adams, R.P. (1995) Identification of Essential Oil Components By Gas Chromatography/Mass Spectroscopy, Allured Publishing Corporation, Carol Stream, IL, USA. Jennings, W. and Shibamoto, T. (1980) Quantitative Analysis of Flavor and Fragrance Volatile by Glass Capillary Column Gas Chromatography, Academic Press, New York, USA. Sadtler Research Laboratories (1984) The Sadtler Standard Gas Chromatography Retention Index Library, Sadtler Research Laboratories, Philadelphia, PA, USA. Websites Name

Location

Reading University – LRI and Odour Database Cornell – Flavornet The Pherobase

http://www.odour.org.uk/information .html http://www.flavornet.org/flavornet.html http://www.pherobase.com/database/ kovats/kovats-index.php http://www.crec.ifas.ufl.edu/rouseff

University of Florida – Citrus Database

can then be compared on the same wax column for possible identity confirmation. LRI data bases can also be found in books but these are costly and in many cases out of print or have used chromatographic stationary phases which are no longer commonly used. Data bases are free, readily available, and much faster at locating data. These books are listed as possible references as well in Table 3.3. 3.3.2.2 GC–MS Gas chromatography-mass spectrometry is an instrumental combination of the separation power of high resolution capillary gas chromatography with the identification power of the mass spectrometer. The combination is a powerful tool in the analysis of complex flavor mixtures and an essential piece of instrumentation in any modern flavor analysis lab. It works on the principle that volatiles are fragmented into ions of predictable size and frequency. The weakest chemical bonds holding the molecule together will be the place at which the molecule is most frequently fragmented and ions form. The most common fragmentation uses electron impact (EI) at about 70 eV. This high energy stream of


65

electrons fragments the volatiles as they elute from the end of the capillary GC. The ions formed are then focused and sent to a mass analyzer that sorts the ions in terms of their mass to charge ratio, abbreviated m/z. The numbers of ions at each mass are counted and displayed as a bar chart diagram for each mass. The resulting fragmentation pattern is called a mass spectrum and is characteristic of each molecule. The mass spectrum can be compared with other mass spectra either self-determined or from commercial sources such as John Wiley & Sons, Inc. and NIST (National Institute of Standards and Technology). Unfortunately, these libraries were created for environmental, hospital, and forensic labs and the vast majority of their contents are of little interest to flavor analysts. John Wiley & Sons, Inc. has recently introduced a flavor and fragrance library that has addressed this problem. In addition, Allured Publishing has a small essential oil library that is based primarily on ion trap MS. Modern computer based pattern recognition programs match the spectrum obtained with those found in their data base and lists the identifications of those with the highest match. Misidentifications are still possible as some compounds (especially terpenes) have very similar spectra. Therefore, retention index values are also employed to confirm the identity of the chromatographic peak of interest. The Allured essential oil MS library includes DB-5 retention index (LRI) values for each volatile. Beginning in 2005, NIST libraries have begun to include retention index values for some but not all of the volatiles in its libraries for several of the major chromatographic stationary phases. The matching of MS fragmentation pattern and LRI values with standards are considered solid identification confirmation. If only library spectra and literature LRI values are matched the identification is generally considered tentative. 3.3.2.3 MS/MS Shown in Figure 3.5 is a portion of a TIC chromatogram from a grapefruit juice extract. It illustrates the complexity and wide concentration ranges found in many flavor samples. Of interest is the small peak labelled 46. This peak is due to vanillin which produces a strong aroma peak (in GC/O) even though its TIC peak is very small and poorly resolved. Generally, when increased sensitivity and selectivity are necessary, the MS is switched to the selected ion (SIM) mode. Instead of scanning from the customary 25–300 m/z range the MS is set at only one or two masses. In this way a hundred- to thousandfold increase in sensitivity is achieved as the MS collects all the ions from the masses selected instead

66


43 27

18

33 21

30

26

34

44

35

19

39 28

16 20

18

20

49-Caryophyllene

13-Limonene

MS (TIC)

17

45

46

22 23 24 25

22

31 32

29

24

36

41 42

40

46 47 48

37 38

26

28

30

Retention time (min)

Figure 3.5 GC/MS TIC chromatogram of a grapefruit juice extract in chromatographic region between limonene and caryophyllene. The small shoulder peak labelled 46 is due to vanillin and is magnified in the inset for greater clarity.

of just a few in the brief time ion signals are collected in the scan mode. However, it is critical that the masses selected be characteristic and as unique as possible as the trade off for increased sensitivity is the loss of full spectrum identification. If at least two masses are collected then the ratios of these masses in the unknown can be compared to that of the standard as at least a partial confirmation that the ions collected at the appropriate retention time are only due to those of the compound of interest. Although not as widely available (due to increased instrument costs) is the use of GC/MS/MS. In this situation the first MS acts as a filter to allow only a single mass (often the molecular ion, M+ ) of interest to proceed to the second MS. It is ionized prior to entering the second MS and a full mass spectrum of the compound of interest can be obtained. Shown in Figure 3.6 is an example for the case of vanillin in a grapefruit juice extract. In comparing Figures 3.5 and 3.6, it should be noted that the use of MS/MS detection allows for shorter run times as the chromatographic separation need not be as good because the first MS filters out all but the mass of the compound of interest. In the case of vanillin, the first MS was set at 152 (the molecular ion). Then only the fragments from this mass are recorded in the MS/MS mode. It can be seen that the baseline drops sharply due to the filtering effect of the first MS. The peak at 8.3 min in Figure 3.6 was confirmed as vanillin by comparing its spectra with that of standard vanillin shown

67

Intensity of m/z

300

MS/MS OFF

Intensity

MS/MS ON


Purtty:765 fit: 765 R-fit: 780

Sample

Standard

200 40

60

80 100 120 mass (m/z)

140

100

0 7.5

8.0

8.5 Retention time (min)

9.0

9.5

Figure 3.6 Selective MS/MS detection and quantitation of vanillin in a grapefruit juice using m/z 123 which is the major ion in vanillin. From [13].

in the insert. GC/MS/MS is a very powerful separation technique that allows for shorter analysis time with increased sensitivity and selectivity. The only drawbacks are higher instrument costs and MS/MS spectra are sufficiently different from standard GC/MS spectra that all confirming spectra must be self-generated from standards.

REFERENCES 1. Linner, R. T. (1970) Caramel color: a new method of determining its color hue and tinctorial power, in Proceedings of the Society of Soft Drink Technologists Annual Meeting, pp. 63–72. 2. Schanda, J. (2007) Colorimetry, Wiley & Sons, Inc., New York, USA. 3. Nielsen, D.M. and Nielsen, G.L. (2006) Ground-water sampling, in Practical Handbook of Environmental Site Characterization and Ground-Water Monitoring (ed. D. M. Nielsen), Taylor and Francis Group, CRC Press. Boca Raton. USA, pp. 959–1112. 4. Fennema, O. R. (ed.) (1985) Food Chemistry, 2nd edn revised and expanded, Marcell Dekker, Inc., New York, USA, pp. 46–50. 5. Scholz, E. (1984) Karl Fischer titration: determination of water (chemical laboratory practice), Springer, Germany. 6. Hecht, E. (1998) Optics, 3rd edn, Addison-Wesley, New York, USA. 7. Born, M. and Wolf, E. (2000) Principles of Optics: Electromagnetic Theory of Propagation, Interference and Diffraction of Light, 7th edn, Cambridge University Press, Cambridge, UK.

68


8. Boulton, R., Singleton, V., Bisson, L., and Kunkee, R. (1996) Principles and Practices of Winemaking, Chapman & Hall, UK. 9. Symon, K. (1971) Mechanics, 3rd edn, Addison-Wesley, New York, USA. 10. Kovats, E. (1958) Gas-chromatographische charakterisierung organischer verbindungen.1. Retentionsindices aliphatischer halogenide, alkohole, aldehyde und ketone. Helv. Chim. Acta, 41, 1915–1932. 11. Vandendool, H. and Kratz, P. D. (1963) A generalization of retention index system including linear temperature programmed gas-liquid partition chromatography. J. Chromatogr., 11, 463. 12. Arthur, C. L. and Pawliszyn, J. (1990) Solid phase microextraction with thermal desorption using fused silica optical fibers. Anal. Chem., 62, 2145–2148. 13. Goodner, K. L., Jella, P., and Rouseff, R. L. (2000) Determination of vanillin in orange, grapefruit, tangerine, lemon, and lime juices using gc-olfactometry and gc-ms/ms. J. Agric. Food Chem., 48, 2882– 2886.

4 Gas Chromatography/ Olfactometry (GC/O) Kanjana Mahattanatawee Department of Food Technology, Siam University, Thailand

Russell Rouseff IFAS, Citrus Research and Education Center, University of Florida, USA

4.1 INTRODUCTION Gas chromatography/olfactometry (GC/O) is a hybrid technique that combines the separating power of gas chromatography (GC) with the specific selectivity and sensitivity of the human nose. Gas chromatography is a relatively mature science and dozens of excellent books have been written about this topic during the last 50 years. This chapter will deal primarily with the olfactometry aspects of GC/O as it would be beyond the scope of this chapter to also discuss the many aspects of gas chromatography. GC/O is a method to determine directly which compounds in a complex mixture of volatiles have aroma activity. Although chromatographers had noted specific aromas eluting during GC runs, it Practical Analysis of Flavor and Fragrance Materials, First Edition. Edited by Kevin Goodner and Russell Rouseff. © 2011 Blackwell Publishing Ltd. Published 2011 by Blackwell Publishing Ltd.

70


was not until 1964 that it was proposed that a dedicated human evaluate the eluting aroma compounds during a GC separation [1]. Although GC/O can determine which volatiles have aroma activity and which do not, it should be kept in mind that GC/O results do not provide an indication of how these aroma-active compounds will interact with each other or with the food matrix. Preliminary identifications of these aroma-active compounds are usually achieved by comparing retention behavior and sensory descriptors from 2–3 dissimilar capillary GC column types. Tentative identifications can be confirmed by matching the retention times and sensory characteristics of authentic standards with that from the tentatively identified volatile. Whenever possible, mass spectrometry is also employed as an identification confirmation tool. Minimum requirements for the proper identification of GC/O aroma volatiles have recently been published [2] and selected aspects will be discussed in more detail in later sections of this chapter. Traditional chromatograms of food volatiles employing either FID or MS detection will indicate which volatiles are present in highest concentration, as both FID and MS detectors are general mass detectors. Unfortunately FID or TIC chromatograms usually do not represent the aroma profile of the food because most aroma active compounds exist as potent, low concentration volatiles which produce little FID or MS response while other compounds exist in high concentrations with little or no aroma activity. The human nose is a highly selective and highly sensitive detector which has a theoretical odor detection limit of about 10−19 moles [3]. Therefore it is the detector of choice when determining aroma activity. Since the limit of detection of the human nose is lower than most instrumental detectors for many compounds, it is often necessary to concentrate the sample in order to identify specific volatiles using instrumental techniques.

4.2 ODOR ASSESSORS’ SELECTION AND TRAINING Although many of the early GC/O studies were conducted using a single trained odor evaluator (sometimes called a sniffer), two to three trained sniffers are now considered an absolute minimum. Multiple sniffers are needed to compensate for individual threshold differences and to eliminate or at least minimize problems due to specific anosmias. Unfortunately, there is no commonly accepted training procedure for GC/O sniff panel members and there are conflicting reports about the benefits of assessor training [4–6]. Friedrich and coworkers [7] have

GAS CHROMATOGRAPHY/OLFACTOMETRY (GC/O)

71

suggested that a standard set of 40 odorants would be sufficient to cover all aroma categories. Normally a set of standards containing 10 to 20 of the anticipated aroma-active compounds is used for training. Panelists would typically be expected to evaluate this set of standards repeatedly until they could consistently respond to the aroma compounds of interest. Panelists who cannot perform consistently (reproducibly) or cannot detect certain of the training compounds should not be used. Training normally takes at least 3–6 weeks unless the sniffers have had previous sensory or GC/O experience. However, there are some forms of GC/O such as the frequency of detection method that utilizes 10–12 untrained sniffers. The implicit assumption of this technique is that there will be a normal distribution of aroma thresholds within the panel and they are familiar with whatever technique is used to indicate an aroma response.

4.3 SENSORY VOCABULARY The description of aroma volatiles as they elute from the GC column by sensory assessors can be a useful piece of information in the identification of individual aroma volatiles. Usually investigators will borrow heavily from the vocabulary established from sensory aroma profiling studies. A panelist’s initial description of aroma compounds is based upon the assessor’s experience. Compounds are typically described in terms of previously experienced foods or other volatile substances. There have been limited attempts to standardize aroma vocabulary. It has been complicated by the fact that there are often disagreements in terms of what something smells like because of genetic differences in olfactory receptors. In terms of practical GC/O, two basic approaches have been developed. One is a fixed choice procedure in which a list of descriptors is developed from several initial GC/O runs and all subsequent responses must be selected from that list. The other option is to allow free choice descriptions. This option allows for more flexibility in description and takes into consideration perception variations among sensory assessors. This second option also does not require as much training as a fixed list because it does not require agreement in terms of aroma perceived. However, the interpretation of the final data is more difficult because assessors will use different descriptors for the same aroma. Often the only way to resolve the discrepancy in description is to rerun a standard (providing identification can be made) and ask the assessor if that is the same aroma that they observed in the sample.

72


With the fixed choice option data interpretation is less ambiguous as the assessor is forced to choose a particular descriptor. This forced choice requirement can sometimes be confusing to assessors when dealing with complex samples and aroma components which elute rapidly and in close proximity to each other. Assessors can sometimes be searching amongst the list of descriptors and miss the next aroma component.

4.4 GC/OLFACTOMETERS (SNIFFERS) There are a number of olfactometer designs which have been developed to evaluate aroma volatiles as they elute from the end of a GC column. Some units are sold separately (such as the ones from Gerstel (Baltimore, ¨ MD, USA), SGE (Austin, TX, USA) and Brechbuhler (Houston, TX, USA)). Others are sold only as part of a complete instrument (such as the ones from Datu, Geneva, NY, USA, and Microanalytics, Round Rock, TX, USA). They all share a single feature. That is, they all allow for the provision of humidified air to be added to the effluent stream from the GC column. Some designs also allow for the addition of a neutral make-up gas to change the overall velocity of the gas mixture as it impinges on the assessor’s nose. One study reported that as overall gas velocity increased aroma detection accuracy increased [8]. However, this study employed a particular olfactometer design which may or may not be applicable to other olfactometers. The purpose of the humidified air is to cool the heated gases (as much as 250 ◦ C) as they elute from the GC column oven and to prevent dehydration of the assessor’s nasal passages during sniffing. The relative amount of this cooled humidified air varies considerably among the different designs. Another difference between olfactometer designs is how close to the nose the humidified air is added to the GC column effluent. In some cases, the humidified air is added directly into the sniffing cone only a few centimeters from the assessor’s nose. Other designs add the humidified air more than 50 cm ahead of the assessor’s nose. A final and potentially more important design factor is the distance between the end of the column and the assessor’s nose. This distance will impact on how much mixing the GC effluent will undergo before it is assessed. In some designs this distance is 2 cm. Other designs use the base of the FID detector to introduce the effluent into the fast moving stream of humidified air in a 1 cm diameter tube about 80 cm long. An important factor to consider in the use of GC/olfactometers is the relative comfort of the assessor during the sensory evaluation. A general


73

rule of thumb is to adjust the chromatographic conditions such that the evaluation time would not exceed 30 minutes. Some designs require the assessor to stand or assume positions close to the GC that are uncomfortable or may distract from the assessors ability to focus on aroma evaluation simply because of physical discomfort. Ideally, the assessors should be seated in a comfortable position so that they can be alert and focus their attention entirely towards aroma detection and description.

4.5 PRACTICAL CONSIDERATIONS There are a few practical factors well known to experienced workers in the area of GC/olfactometry, but not immediately obvious to workers just entering this field. First of all the GC/O unit should be situated in a low traffic area to minimize assessor distraction. It should also be situated away from any food preparation/cooking areas such as microwave ovens or kitchens. The instrument should be located in a separate room equipped with charcoal filter ventilation and at positive pressure compared to the surrounding rooms. Finally, assessors should not use colognes, scented deodorants or hair sprays on the days they will be assessing GC peak odors. The assessors should not have eaten or have drunk a flavored beverage for at least 1 hour before sniffing. Many of these later suggestions are identical with common sensory practices.

4.6 TYPES OF GC/OLFACTOMETRY Four different GC/O techniques have been developed to identify potent odorants in foods. They include: dilution analysis, time-intensity methods, detection frequency methods, and posterior intensity methods.

4.6.1 Dilution Analysis This is the oldest and most popular GC/O method. It determines the relative strength of the aroma-active volatiles based on stepwise dilution, usually as a series of 1:2 or 1:3 dilutions. Starting with the most concentrated sample and proceeding to successively more dilute samples, each dilution is sniffed until no odor is detected. However, the diluted samples should be presented in randomized order to avoid bias introduced by knowledge of the samples. The most diluted sample in which an

74


odor-active compound can detected is called a dilution value or dilution factor value (FD value). For example, if the original aroma extract was diluted as a series of 1:2 (one part in two parts of solvent), and the odor of interest was perceived until the sixth dilution, then the dilution value was 25 = 32. All forms of dilution analysis require the aroma volatiles to be concentrated in some way. Solvent extraction is the most common procedure used to extract and concentrate aroma volatiles for this type of GC/O. Two slightly different dilution techniques have been developed by research groups in America and Germany: charm analysis by Acree and coworkers [9] and aroma extract dilution analysis (AEDA) by Grosch and coworkers [10, 11]. These methods are based on odor detection thresholds rather than psychological estimations of stimulus intensity at super threshold levels. The difference between the two methods is that charm analysis integrates and sums the dilution values and odor response durations, whereas AEDA determines only maximum dilution value. Charm analysis requires specific software and computer availability where as AEDA can be done with only pen and paper, and is perhaps the reason why it is more widely used. An AEDA aromagram can be obtained by plotting maximum dilution value on the y axis and retention time (or retention index) on the x axis. An example of an AEDA aromagram is presented in Figure 4.1 demonstrating the aroma components in an extract from freshly prepared popcorn [12]. This method has been employed to determine the potent odor-active compounds in many food products including (but not limited to): grapefruit juice [13], green and black tea [14], cheese [15], and coffee [16]. In charm analysis, when the sniffer detects an odor in the olfactometer air stream, they press the mouse button and hold it for the duration a particular odor is perceived. When the odor is no longer perceived the sniffer releases the mouse button and indicates the odor character from a predetermined list of sensory descriptors. Sniffers must be trained and must have developed and mastered a suitable vocabulary or descriptor list. Times and durations of the individual sniffs are combined and graphed to yield an aromagram with peaks and integrated peak areas (charm values), which are used to quantify potency. A charm value can be calculated according to the formula c = dn−1 where n is the number of coincident responses and d is the dilution factor [17]; Figure 4.2 shows an idealized construction of a charm aromagram. Charm aromagrams are generated when the time-dilution strength blocks produced from all the samples in a dilution series are added together to produce a

75

GAS CHROMATOGRAPHY/OLFACTOMETRY (GC/O) 20

3

512

19

2

256

FD-factor

13

6

128 5

64

8

17

9

32

7 10 11

16 1

8

800

900

22 23

18 1415

4

1000

21

16

12

1100 1200 (RI, SE 54)

1300

1400

Figure 4.1 Dilution factor (FD) chromatogram of the odorants of an extract from freshly prepared popcorn where 1 = methional, 2 = 2-furfurylthiol, 3 = 2-acetyl-1-pyrroline, 4 = 1-octen-2-one, 5 = 2-propionyl-1-pyrroline, 6 = 2acetyltetrahydropyridine, 7 = Furaneol, 8 = 2, 5-dimethyl-3-ethylpyrazine, 9 = 2-methoxyphenol, 10 = unknown, 11 = 2-acetyltetrahydropyridine, 12 = (Z)-2nonenal, 13 = 2, 3-diethyl-5-methylpyrazine, 14 = unknown, 15 = 2, 4-nonadienal, 16 = (E,E)-2,4-nonadienal, 17 = unknown, 18 = (E,Z)-2,4-decadienal, 19 = 4vinyl-2-methoxyphenol, 20 = (E,E)-2,4-decadienal, 21 = 4, 5-epoxy-(E)-2-decenal, 22 = (E)-β-damascenone, 23 = vanillin [12].

coincident response chromatogram (see Figure 4.2). To make the peak areas proportional to the amount of compound present, the cumulative responses must be transformed using the algorithm shown in Equations (4.1) and (4.2): dv = Fn−1 di charm = dv

(4.1) (4.2)

peak

where dv is the dilution value (or number of dilutions made before a particular dilution becomes odorless), F is the dilution factor that was used to prepare the dilution series, n is the number of dilutions in the series that yielded a square block at a particular chromatographic elution time, and di is the time or retention index over which the responses are determined. The software program then combines the data from several sniffs using Equation (4.1) to produce a chromatogram. The

76

PRACTICAL ANALYSIS OF FLAVOR AND FRAGRANCE MATERIALS Sniff port responses Dilution 1 Dilution 2 Dilution 3

Coincident responses chromatogram

n 3 2 1 0

Dilution value

9

F n−1 Charm chromatogram F = Dilution factor Charm = Peak area

3 1 0 Retention index

Figure 4.2 Construction of a charm chromatogram [17].

chromatogram is integrated in the region of a peak using Equation (4.2) to yield the charm value for each aroma-active peak. Charm analysis has been employed for the determination of odoractive compounds in a wide range of foods, beverages, and spices such as: coffee [18, 19], stored boiled potatoes [20], coriander [21], beer [22], and citrus peel oil [23].

4.6.2 Time Intensity McDaniel and coworkers [24–26] developed a time-intensity GC/O method called OSME (derived from the Greek word for smell) which was based on psychophysical laws. Psychophysical smell behaviors are most commonly represented by Stevens’ Law [27, 28]. Stevens established that the odor intensity (I) of a compound increases with its concentration (C) that exceeds its threshold raised to a power n: I = k(C − T)n

(4.3)

where T is the compound’s threshold value and k is the constant of proportionality. Furthermore, under Stevens’ Law, two different

77


Saturation

Response

A’ A B B’ Threshold

0

10

100

1000 Dose

10 000

100 000

Figure 4.3 Stevens’ Law dose response curves for two substances with the same threshold but different dose response behaviors. At 10× concentrations the responses A and B are appreciably different even though they would have identical odor activity values (OAVs).

compounds at the same concentration (C), and possessing very close threshold values (T), but showing different exponents (n), may provide different individual odor intensities (I) (Figure 4.3) and consequently provide different individual odor contributions to the intensity and quality of a flavor system. The OSME method differs from charm analysis and AEDA in that only a single concentration of the extract is evaluated. No dilution series are evaluated; hence OSME is not based on odor detection thresholds but on perceived odor intensity. OSME or time–intensity methods are based on estimation of the odor intensity with time for each compound detected at a sniffing port. The panelist moves a variable resistor as a function of the intensity of perception and at the same time describes the odor of the eluting compound. The trained sniffer rates the intensity of an eluting compound using an electronic time–intensity scaling device (15 cm scale, 0 = none, 7 = moderate, and 15 = extreme) coupled with computer data-handling software which provides an FID-style aromagram called an osmegram (Figure 4.4). In this study [26] a trained panel of four assessors was used to determine relationships between odor intensities and concentration in a series of aroma-active standards. Both the maximum odor intensity of the compounds and the area under the odor intensity peak showed significant correlations with the physical concentration of the compounds in the GC effluent.

78

6

Imax

Floral

8

Floral

10

Floral/musty

Green leavels/rubber

Aroma intensity (16-point scale)

12

Citrus

Soapy/floral


4 2

d

0 1200

1300

1400 Kovats indices

1500

1600

Figure 4.4 Osmegram of aroma active standards [26].

Etievant et al. [29] reported a cross-modality matching method with the finger span (GC/O/FSCM) based on the same principle. They described a prototype for the precise measurement and acquisition of the distance between the thumb and another finger during analysis (Figure 4.5). A four-member panel was able to determine intensity (4) (5) (3) (2) (1)

Figure 4.5 Finger span prototype used to measure distance between finger: (1) fixed ring for the thumb; (2) mobile ring for the major or the index finger connected to a 195 mm long rheostat; (3) cursor track; (4) signal lamp; (5) on/off switch [29].

79


characteristics of 11 standard solutions with good precision. However, individual panelist reproducibility varied considerably. This variation between assessors demonstrated the importance of the using a panel rather than a single assessor to evaluate aroma active materials. Time intensity GC/O is not as popular as GC/O dilution techniques because it requires additional hardware (intensity transponder) and a second data channel with corresponding software. It has been used (for example) in hop oil [30], blackberries [31], wines [25], and extensively used in citrus aroma [32–35]. Using chromatographic software such as ChromPerfect allows for the rapid visualization of individual FID and GC/O responses. An example is shown in Figure 4.6, where the GC/O response has been inverted in the so called ‘fishbone’ fashion for improved clarity.

4.6.3 Detection Frequency

Nootkatone

β−Sinensal

Neral

Dodecanal

Geranial Nerol 2E, 4E-Decadienal Geraniol

100

2E-Decenal

Response (mV)

200

Linalool Octanol Terpenen-4-ol

300

Nonanal

Limonene

400

Citronellal

Octanal

500

Decanal

The detection frequency method uses the number of assessors who perceived an odor at the sniff port from a single concentration sample. However, a larger group (six to 12) of untrained assessors is employed to determine odor perception, rather than estimating intensity or noting

0

400

800

1200

6

8

10

12

14

16

20 18 Time (min)

22

24

26

28

30

32

Figure 4.6 GC/FID chromatogram (top) and time-intensity aromagram (inverted, bottom) of a grapefruit oil, separation performed on a DB-Wax column [34].

80


aroma activity in successive dilutions. The percentage of assessors detecting an odor-active compound was reported to correspond with odor intensity [36–39]. The method was first proposed by Linssen et al. [36] using a group of 10 assessors to overcome the limitation of a small number of assessors and only one dilution level was used. The aim of the study was only to identify the volatile compounds that were responsible for the taint found in water packed in LDPE/aluminum/cardboard laminate packaging. Van Ruth et al. [40] evaluated odor released from rehydrated French beans. Ten assessors detected and rated perceived intensities of the eluting compounds at the sniffing port on a nine-point intensity interval scale (1 = extremely weak; 9 = extremely strong). Tenax TA tubes without adsorbed volatile compounds were used as dummy samples for determining the signal-to-noise level of the group of assessors. GC sniffing of dummy samples showed that detection of an odor at the sniffing port by one or two of the 10 assessors can be considered as ‘noise’. The number of assessors perceiving an odor correlated significantly with odor intensity scores (Spearman’s ranked correlation test, P < 0.05) indicating that the number of assessors is a sufficient measure for odor intensity. Furthermore, significant correlations between the number of assessors perceiving odor active compounds correlated well with the intensity scores of attributes in sensory analysis [40–42]. An example of a sniffing chromatogram by detection frequency is shown in Figure 4.7. Pollien et al. [38] have reported a similar technique based on the frequency of detection with six untrained assessors. Elution of odoractive compounds was recorded by pressing a button during the whole sensory impression. The square signal was registered by an HP Pascal workstation. When peak recognitions were needed, the assessor recorded the corresponding odor descriptors on a tape recorder. The six individual aromagrams of a given sample were summed to one chromatogram and normalized with homemade software, yielding an averaged aromagram. Peak heights (percentage detection frequency of an odor by panelists) and areas (expressing the detection frequencies and the detection time duration) are called NIF and SNIF (nasal impact frequency and surface of nasal impact frequency, respectively) (Figure 4.8). It was claimed that the method provided both repeatability and reproducibility. The authors had set an experiment for the number of assessors to establish an aromagram. Details about the procedures may be found in the literature cited above. These results suggest that one or two untrained panelists cannot provide a reliable aroma profile. The first assessor may detect or miss a peak above his/her own detection threshold. Even when the

81

GAS CHROMATOGRAPHY/OLFACTOMETRY (GC/O) 10

4

14

5

9

Number of assessors

8 7 9

13

6 1 3

7

5 11

6 4 2

10

12

3 2 1 0 0

10

20 30 Retention time [min]

40

50

Figure 4.7 Sniffing chromatogram of volatile compounds of diced French beans, rehydrated for 5 min in a closed flask; numbers on the chromatogram refer to identified compounds (data not shown) [40].

second assessor correctly detects the same peak, a 50 % difference from the asymptotic result can be observed. As a result, the reasonable number of untrained assessors should be eight to 10 panelists [38]. There have been subsequent attempts to quantify odorants using the GC/SNIF method where detection frequency is correlated with odorant concentration. It has been suggested that quantification is possible when the odorants have unimodal detection thresholds as a function of the logarithm of their concentration. However, the primary assumption in SNIF quantification is that the panel has a normal distribution of aroma thresholds amongst the assessors for the compounds of interest. This procedure can also be applied to odorants whose concentration is below the sensitivity of physical detectors or when the odorant coelutes in an inseparable mixture of volatiles which have high odor thresholds [39, 43]. The method has been used for analyzing odor active compounds in foods such as: French beans, bell peppers and leeks [44], yogurt [45] mineral water in polyethylene laminated packages [36], black currant [46], and orange juice [39].

82

PRACTICAL ANALYSIS OF FLAVOR AND FRAGRANCE MATERIALS Aromagram = Average of individual aromagrams

Individual aromagrams 1 +

t (min)

NIF % 100

1 75 +

t (min)

50

1 25 +

t (min) t (min)

1

t (min)

Figure 4.8 Data treatment procedure using the responses from four panelists [38].

4.6.4 Posterior Intensity Method The posterior intensity method involves the estimation of the odor intensity of a compound in the GC effluent by trained assessors. The perceived intensity is scored on a scale after a peak has eluted from the GC column. The method has been called gas chromatography/sniffing port analysis (GC/SP) [40], odor-port evaluation or GC/sniffing [47], GC odor profiling [48], posterior intensity [49], or perceived intensity method [50]. They are based on the same principle but differ in the number of assessors [47–49] and intensity scale (3, 5, and 9 point intensity [40, 47, 51] respectively). The posterior intensity method is actually a hybrid method employing aspects of time intensity as well as detection frequency procedures. This method is similar to the detection frequency method in that a relatively large group of assessors is used, but differs in that assessors must estimate intensity rather than simply indicating aroma activity. It is similar to OSME/time intensity in that panelists are required to estimate aroma intensity but do not do so continuously, only after an aroma active peak elutes. Therefore, responses can be recorded with pen and paper and a second computer data channel is not required. The disadvantage is that they are also required to record the time a peak elutes (something done automatically when computer software is employed).


83

To validate the posterior intensity method, Van Ruth et al. [49] examined the relationships between perceived GC/O odor intensities with perceived intensities of the same compounds in sensory headspace analysis. The correlation between GC/O posterior intensity scores and sensory odor intensity scores were highly correlated. The method has been applied in studies on Cheddar cheese [47] rehydrated French beans [40], orange juice [48], and wines [52].

4.7 SAMPLE INTRODUCTION One of the most overlooked and undoubtedly one of the most important aspects of GC/O is the process of collecting and introducing into the GC a representative sample of the product being evaluated. The entire second chapter in this book is devoted to this topic. In the case of dilution analysis, the method of sample preparation will almost always be solvent extraction. Different extracting solvents will produce different aroma profiles because all solvents selectively extract some aroma components more efficiently than others. Therefore the choice of the extracting solvent will greatly influence the aroma components that are introduced into the GC. Traditionally polar solvents such as diethyl ether and ethyl acetate will extract more of the polar aroma components than will nonpolar solvents such as pentane. A serious limitation of this approach is that solvent extractions do not allow for the analysis of highly volatile components as these elute at the same time as the solvent. Another problem with solvent extraction is that it extracts materials that are semivolatile or nonvolatile which may degrade longterm chromatographic performance or produce thermal degradation artifacts in the hot GC injection port. Headspace analysis is the other option that is commonly employed in GC/O. Classic static headspace analyzes the volatiles in a fixed volume of headspace gas and has been used on a wide range of samples with some degree of success. The primary limitation with this technique is that the headspace often contains a large amount of water and relatively small amounts of the components of interest. Therefore the sensitivity from the direct injection of headspace gas is fairly low. Dynamic headspace can be used in those situations where the interest is primarily in the most volatile compounds in the sample. In passing a fixed amount of the gas over the surface of the sample increased amounts of the more volatile compounds can be trapped and later eluted into the GC. An increasingly popular method of sample concentration and introduction is the use of solid phase microextraction commonly known

84


as a SPME. Headspace volatiles are concentrated on the thin layer of absorbent on the silica fiber. The amounts of volatiles collected will be dependent on the temperature of the sample and headspace as well as the exposure time. Different absorbent materials will selectively concentrate headspace volatiles in a manner similar to solvents of different polarity in solvent extraction. Therefore the choice of solid phase in adsorbent will greatly influence the relative amounts of volatiles observed. The guiding principle should always be to choose a fiber that will produce the most representative sample of the headspace gases.

4.8 IDENTIFICATION OF AROMA-ACTIVE PEAKS One of the factors that differentiate GC/O from the earlier odoractivity value (OAV) approach is that all aroma-active components could be detected and their intensities recorded without actually having an analytical method to measure them. Therefore, the identity of each volatile measured was known. With GC/O, only the aroma intensity or dilution strength is known along with sensory descriptor. Therefore, the major challenge in GC/O is to identify correctly the volatiles responsible for producing the aroma activity. Correct identification of aroma-active peaks can be extremely challenging as a given food product will contain hundreds of volatiles. Correct identification is further complicated by the fact that food volatiles will differ in concentration by several magnitudes. The disadvantage of the GC/O approach is that aroma-active peaks must be identified using a variety of techniques and usually confirmed with some analytical method. Tentative identifications can be based on limited GC/O data. However, firm identifications require sensory and analytical confirmations from at least two independent methods.

4.8.1 Standardized Retention Index Values One of the primary ways that aroma-active peaks are identified (at least in a preliminary fashion) is by using either the alkane- or esterbased standardized retention index systems. The reporting of retention times is unacceptable in that these values have little value outside of the reporting lab as they are subject to flow rate, oven temperature program, and column length differences. The alkane-based system is more commonly known as the linear retention index or LRI system. In this system a progressive series of linear alkane’s is used to standardize all the aroma volatiles. In practice, this requires a sample containing a


85

series of linear alkanes from C 5 to C 20 (sometimes as high as C 25) to be injected under the same conditions as samples will be run under.

4.8.2 Aroma Description Matching Tentative identifications can often be achieved by matching the retention values on two columns and aroma descriptions from standardized tables found on the web. However, the lack of a standardized vocabulary makes this process difficult as the same compound may be described in a wide range of terms usually based on prior experience.

4.8.3 MS Identifications Mass spectrometry (MS) is a mature technology that has been coupled to gas chromatography for identification purposes. Today these instruments are tuned and calibrated under computer control so that they can now be used by any scientists whereas before a dedicated specialist was required to operate them. Identifications are based on the fragmentation patterns of the compounds as they elute from the end of the GC column. The fragmentation patterns are highly reproducible and characteristic of each compound. Libraries containing over 100 000 volatiles can be purchased from NIST or John Wiley & Sons, Inc. Compounds are identified by matching the mass of fragments and relative intensity with those in the library using one of several computer pattern matching programs. This has allowed chromatographers to identify components separated by the GC column rapidly. However, there has been an overreliance on the computer-derived matching identification. Problems occur when several compounds produce similar fragmentation patterns, for example, terpenes. Other problems occur when the fragmentation pattern of the compound of interest is not present in the library. When this happens the computer will identify the compound incorrectly with the closest match it can find in its library. The need to couple MS pattern matching identification with some independent data such as retention index values has been discussed [2]. The most recent version of the NIST library has included retention index values for many compounds to help users select the correct identification from the list of matches proposed by the computer. The use of MS to identify aroma compounds can be extremely challenging when the aroma volatile is extremely potent and coelutes with an inactive volatile present in much greater concentrations. In this case

86


the MS will correctly identify the volatile present in highest concentration, but it may not be the aroma-active compound. Often this false identification can be ruled out from aroma descriptions or more reliably from retention characteristics on two or more columns. In the final analysis, overreliance on MS identifications should be avoided and correctly considered as only one piece of information which must be confirmed with additional independent data.

4.8.4 Use of Authentic Standards After a preliminary identification has been made by matching several pieces of independent data from the literature or internal databases, the identification should be confirmed using authentic standards. The standards should be run under the same conditions as the samples and the data compared. LRI values on polar and nonpolar columns should match within 1 %. Mass spectrometry fragmentation patterns should be essentially identical except for compounds found in low concentration producing signal to noise ratios less than 10:1. The matching of sensory description of the standard compared with that found in the sample at the same retention time is also highly desirable, but in order to evaluate properly the concentration of the standard should be matched closely with that found in the sample.

4.9 CONCLUSION GC/O is a powerful technique that has become a staple tool in most flavor chemists’ laboratory. The results from GC/O should not be overinterpreted. It should be remembered that this technique determines the relative aroma activity and aroma character of component volatiles as individual components. Therefore, the comparison of sensory data, where volatiles are evaluated simultaneously in mixtures, with GC/O data – where components are determined individually – should be done with some caution. Finally, it should be recognized that each of the many forms of GC/O have their specific strengths and limitations. All GC/O data will be influenced by the manner by which volatiles are extracted and concentrated, and probably represent the largest source of variation. Another factor to consider is how concentrated should the sample extract be. There is always the possibility of overconcentrating sample volatiles. Time-intensity GC–O has the advantage of sensing aroma


87

volatiles at supra threshold levels but is limited by the wide range that various assessors estimate aroma intensity. This can be minimized by normalizing each assessor’s data. Dilution methods are based on threshold information with the assumption that the dose response slopes for all volatiles are identical or very similar. There are many cases where this is not true. On the other hand, the data based on threshold characterization tends to be more reproducible than techniques which require magnitude estimation.

REFERENCES 1. Fuller, G. H., Tisserand, G. A., and Steltenk. R. (1964) Gas chromatograph with human sensor – perfumer model. Ann. N. Y. Acad. Sci., 116, 711– 714. 2. Molyneux, R. J., and Schieberle, P. (2007) Compound identification: a Journal of Agricultural and Food Chemistry perspective. J. Agric. Food Chem., 55, 4625– 4629. 3. Reineccius, G. (1994) Source Book of Flavors, 2nd edn, Springer, New York, USA p. 24. 4. Hulin-Bertaud, S., O’Riordan, P. J., and Delahunty, C. M. (2001) Gas chromatography/olfactometry panel training: a time-intensity approach for odor evaluation, in Food Flavors and Chemistry (eds A. M. Spanier, F. Shahidi, T. H. Parliment, and C.-T. Ho) Royal Society of Chemistry, London, UK, Vol. 274, pp. 394–403. 5. Callement, G. et al. (2001) Odor intensity measurements in gas chromatographyolfactometry using cross modality matching: evaluation of training effects. ACS Symp. Ser., 782, (Gas Chromatography-Olfactometry), 172–186. 6. Van Ruth, S. M. (2001) Methods for gas chromatography-olfactometry: a review. Biomolecular Engineering, 17, 121–128. 7. Friedrich, J. E., Acree, T. E., and Lavin, E. H. (2001) Selecting standards for gas chromatography-olfactometry. ACS Symp. Ser., 782, (Gas ChromatographyOlfactometry), 148–155. 8. Hanaoka, K., Sieffermann, J.-M., and Giampaoli, P. (2000) Effects of the sniffing port air makeup in gas chromatography-olfactometry. J. Agric. Food Chem., 48, 2368– 2371. 9. Acree, T. E., Barnard, J., and Cunningham, D. G. (1984) A procedure for the sensory analysis of gas chromatographic effluents. Food Chem., 14, 273–286. 10. Schieberle, P. and Grosch, W. (1987) Evaluation of the flavor of wheat and rye 111– 113. 11. Ullrich, F. and Grosch, W. (1987) Identification of the most intense volatile flavor compounds formed during autoxidation of linoleic acid. Z. Lebensm.-Unters. Forsch., 184, 277–282. 12. Schieberle, P. (1991) Primary odorants in popcorn. J. Agric. Food Chem., 39, 1141– 1144. 13. Buettner, A. and Schieberle, P. (1999) Characterization of the most odor-active volatiles in fresh, hand-squeezed juice of grapefruit (citrus paradisi macfayden). J. Agric. Food Chem., 47, 5189– 5193.

88


14. Guth, H. and Grosch, W. (1993) Identification of potent odorants in static headspace samples of green and black tea powders on the basis of aroma extract dilution analysis (AEDA). Flav. Fragr. J., 8, 173–178. 15. Qian, M. and Reineccius, G. (2003) Static headspace and aroma extract dilution analysis of Parmigiano Reggiano cheese. J. Food Sci., 68, 794– 798. 16. Blank, I., Sen, A., and Grosch, W. (1992) Potent odorants of the roasted powder and brew of Arabica coffee. Z. Lebensm.-Unters. Forsch., 195, 239– 245. 17. Acree, T. E. (1993) Bioassays for flavor, in Flavor Science: Sensible Prinicples and Techniques (eds T. E. Acree and R. Teranishi) American Chemical Society, Washington, DC, USA, pp. 1–20. 18. Sagara, Y. et al. (2005) Characteristic evaluation for volatile components of soluble coffee depending on freeze-drying conditions. Drying Technology, 23, 2185–2196. 19. Deibler, K. D., Acree, T. E., and Lavin, E. H. (1998) Aroma analysis of coffee brew by gas chromatography-olfactometry, in Food Flavors: Formation, Analysis, and Packaging Influences (eds Contis, E. T. et al.), Elsevier, New York, USA, Vol. 40, pp. 69–78. 20. Jensen, K., Acree, T. E., and Poll, L. (2000) Identification of odour-active aroma compounds in stored boiled potatoes, in Frontiers of Flavour Science (eds ¨ Lebensmittelchemie: P. Schieberle, K.-H. Engel), Deutsche Forschungsanstalt fur Munich, pp. 65–68. 21. Eyres, G. et al. (2005) Identification of character-impact odorants in coriander and wild coriander leaves using gas chromatography-olfactometry (GCO) and comprehensive two-dimensional gas chromatography-time-of-flight mass spectrometry (GC ∗ GC-TOFMS). J. Sep. Sci., 28, 1061– 1074. 22. Murakami, A. A. et al. (2003) Investigation of beer flavor by gas chromatographyolfactometry. J. Am. Soc. Brewing Chem., 61, 23–32. 23. Gaffney, B. M., Haverkotte, M., Jacobs, B., and Costa, L. (1996) Charm analysis of two Citrus sinensis peel oil volatiles. Perfumer & Flavorist, 21, 1-2, 4–5. 24. McDaniel, M. R. et al. (1990) Pinot noir aroma: a sensory/gas chromatographic approach. Dev. Food Sci., 24, (Flavors Off-Flavors ’89), 23–36. 25. Miranda-Lopez, R., Libbey, L. M., Watson, B. T., and McDaniel, M. R. (1992) Odor analysis of Pinot noir wines from grapes of different maturities by a gas chromatography-olfactometry technique (Osme). J. Food Sci., 57, 985– 993, 1019. 26. da Silva, M. A. A. P., Lundahl, D. S., and McDaniel, M. R. (1994) The capability and psychophysics of Osme: a new GC-olfactometry technique, in Trends in Food Science (ed. H. M. a. D. G. V. d. Heij) Amsterdam, Holland Elsevier Science B.V. Vol. 35, pp. 191– 209. 27. Stevens, S. S. (1957) On the psychophysical law. Psychological Review 64, 153–181. 28. Stevens, S. S. (1961) To honor Fechner and repeal his law. Science, 133, 80–86. 29. Etievant, P. X. et al. (1999) Odor intensity evaluation in gas chromatographyolfactometry by finger span method. J. Agric. Food Chem., 47, 1673– 1680. 30. Sanchez, N. B. et al. (1992) Sensory and analytical evaluation of hop oil oxygenated fractions. Dev. Food Sci., 29, (Food Sci. Hum. Nutr.), 371–402. 31. Klesk, K. and Qian, M. (2003) Preliminary aroma comparison of Marion (Rubus spp. hyb) and Evergreen (R. laciniatus L.) blackberries by dynamic headspace/OSME technique. J. Food Sci., 68, 697–700.


89

32. Bazemore, R., Goodner, K., and Rouseff, R. (1999) Volatiles from unpasteurized and excessively heated orange juice analyzed with solid phase microextraction and GC-olfactometry. J. Food Sci., 64, 800–803. 33. Goodner, K. L., Jella, P., and Rouseff, R. L. (2000) Determination of vanillin in orange, grapefruit, tangerine, lemon, and lime juices using GC-olfactometry and GC-MS/MS. J. Agric. Food Chem., 48, 2882– 2886. 34. Lin, J. and Rouseff, R. L. (2001) Characterization of aroma-impact compounds in cold-pressed grapefruit oil using time-intensity GC-olfactometry and GC-MS. Flav. Fragr. J., 16, 457– 463. 35. Mahattanatawee, K., Rouseff, R., Valim, M. F., and Naim, M. (2005) Identification and aroma impact of norisoprenoids in orange juice. J. Agric. Food Chem., 53, 393– 397. 36. Linssen, J. P. H., Janssens, J. L. G. M., Roozen, J. P., and Posthumus, M. A. (1993) Combined gas chromatography and sniffing port analysis of volatile compounds of mineral water packed in polyethylene laminated packages. Food Chem., 46, 367– 371. 37. vanRuth, S. M., Roozen, J. P., and Cozijinsen, J. L. (1996) Gas chromatography/sniffing port analysis evaluated for aroma release from rehydrated French beans (Phaseolus vulgaris). Food Chem., 56, 343–346. 38. Pollien, P. et al. (1997) Hyphenated headspace-gas chromatography-sniffing technique: screening of impact odorants and quantitative aromagram comparisons. J. Agric. Food Chem., 45, 2630– 2637. 39. Bezman, Y., Rouseff, R. L., and Naim, M. (2001) 2-Methyl-3-furanthiol and methional are possible off-flavors in stored orange juice: aroma-similarity, NIF/SNIF GC-O, and GC analyses. J. Agric. Food Chem., 49, 5425– 5432. 40. Van Ruth, S. M., Roozen, J. P., Hollmann, M. E., and Posthumus, M. A. (1996) Instrumental and sensory analysis of the flavor of French beans (Phaseolus vulgaris) after different rehydration conditions. Z. Lebensm.-Unters. Forsch., 203, 7–13. 41. Van Ruth, S. M., Roozen, J. P., Cozijnsen, J. L., and Posthumus, M. A. (1995) Volatile compounds of rehydrated French beans, bell peppers and leeks. Part II. Gas chromatography/sniffing port analysis and sensory evaluation. Food Chem., 54, 1–7. 42. Van Ruth, S. M., Roozen, J. P., and Posthumus, M. A. (1995) Instrumental and sensory evaluation of the flavor of dried French beans (Phaseolus vulgaris) influenced by storage conditions. J. Sci. Food Agric. 69, 393– 401. 43. Pollien, P., Fay, L. B., Baumgartner, M., and Chaintreau, A. (1999) First attempt of odorant quantitation using gas chromatography-olfactometry. Anal. Chem., 71, 5391– 5397. 44. Van Ruth, S. M., Roozen, J. P., and Cozijnsen, J. L. (1995) Volatile compounds of rehydrated French beans, bell peppers and leeks. Part 1. Flavor release in the mouth and in three mouth model systems. Food Chem., 53, 15–22. 45. Ott, A., Fay, L. B., and Chaintreau, A. (1997) Determination and origin of the aroma impact compounds of yogurt flavor. J. Agric. Food Chem., 45, 850– 858. 46. Varming, C., Petersen, M. A., and Poll, L. (2004) Comparison of isolation methods for the determination of important aroma compounds in black currant (Ribes nigrum L.) juice, using nasal impact frequency profiling. J. Agric. Food Chem., 52, 1647– 1652.

90


47. Arora, G., Cormier, F., and Lee, B. (1995) Analysis of odor-active volatiles in Cheddar cheese headspace by multidimensional GC/MS/sniffing. J. Agric. Food Chem., 43, 748– 752. 48. Tonder, D., Petersen, M. A., Poll, L., and Olsen, C. E. (1998) Discrimination between freshly made and stored reconstituted orange juice using GC odor profiling and aroma values. Food Chem., 61, 223–229. 49. Van Ruth, S. M. and O’Connor, C. H. (2001) Evaluation of three gas chromatography-olfactometry methods: comparison of odor intensity-concentration relationships of eight volatile compounds with sensory headspace data. Food Chem., 74, 341–347. 50. Van Ruth, S. M. (2004) Evaluation of two gas chromatography-olfactometry methods: the detection frequency and perceived intensity method. Journal of Chromatography, A, 1054, 33–37. 51. Pet’ka, J., Ferreira, V., and Cacho, J. (2005) Posterior evaluation of odour intensity in gas chromatography-olfactometry: comparison of methods for calculation of panel intensity and their consequences. Flav. Fragr. J., 20, 278– 287. 52. Cullere, L., Escudero, A., Cacho, J., and Ferreira, V. (2004) Gas chromatographyolfactometry and chemical quantitative study of the aroma of six premium quality Spanish aged red wines. J. Agric. Food Chem., 52, 1653–1660.

5 Multivariate Techniques Vanessa Kinton Alcohol and Tobacco Tax and Trade Bureau (TTB), Ammendale, USA

5.1 INTRODUCTION Nowadays, some commercially available instruments have chemometric software bundled with their data analysis software. Chemometrics is not typically taught in undergraduate chemical degrees and can intimidate the novice analyst, but chemometric analysis can be a powerful tool for analyzing complex data. The International Chemometrics Society defines chemometrics as ‘the chemical discipline that uses mathematical and statistical methods, (a) to design or select optimal measurement procedures and experiments; and (b) to provide maximum chemical information by analyzing chemical data’ [1]. The usage of chemometrics has increased due to the availability of analytical instruments coupled with fast computers, such as spectrometers and chromatographs that can easily generate megabytes of data. Whereas years ago the computer time required for multivariate analysis of such large data-sets was prohibitive, it is now possible to perform complex statistical analysis and obtain insightful information by applying different multivariate tools in a few minutes. Practical Analysis of Flavor and Fragrance Materials, First Edition. Edited by Kevin Goodner and Russell Rouseff. © 2011 Blackwell Publishing Ltd. Published 2011 by Blackwell Publishing Ltd.

92


Chemometrics can play an important role in the development of an analytical method by assisting the analyst in selecting what variables of an analytical signal are relevant for measurement. Output from instruments that produce sufficiently complex multidimensional signals is often difficult to interpret without the assistance of automated data processing. Some mathematical manipulations of data (averaging, smoothing, spectral analysis via the Fourier transform, etc.) have become common in the software provided with some instruments, but the use of multivariate statistics for recognizing patterns in data and classification of samples by their spectra is less common. Quantitative and qualitative analysis of laboratory data have been largely based on univariate data-sets in which a single variable is used for prediction, quantification, or classification. Choosing the specific wavelengths or retention times for an analyte of interest can be a difficult task, especially in the presence of complex matrices. For example, even when the absorption or emission characteristic of an analyte occurs at a specific wavelength, impurities may be present which compromise the signal of interest. Multivariate techniques allow the analyst to treat the entire spectrum or chromatogram as a pattern, regardless of dimensionality, and to model specific variability due to a chosen analyte in the presence of other substances [2]. In some other instances, the analyst may be interested in comparing an adulterated sample to a reference or a new prototype flavor found in nature to a new formulation developed in the laboratory. This comparison may involve multiple analytical techniques and the data can be very complex. This task, however, can be simplified with the use of chemometrics. For example, comparison of multiple chromatograms can be reduced to a single plot. This chapter describes the basics of multivariate analysis and serves as an introduction to chemometric techniques. In order to perform a multivariate analysis, the analyst should be familiar with some basic notation described in the literature. For example, matrices are usually denoted with upper case bold letters such as: ⎡ ⎤ x11 .. .. x1j .. .. x1p ⎢ : : : ⎥ ⎢ ⎥ ⎢ Xn×p = ⎢ xi1 .. .. xij .. .. xip ⎥ (5.1) ⎥ ⎣ : : : ⎦ xn1 .. .. xnj .. .. xnp where each element xij represents a single object obtained for sample i at variable j. The objects will always be arranged in n rows, where

93

MULTIVARIATE TECHNIQUES

n is the total number of objects (e.g., spectra or chromatograms). The variables (measurements) will be in p columns, where p is the total number of variables or features (chromatographic peaks, wavelengths, etc.). The use of spreadsheets, such as Excel, makes it easier for the user to understand the concept of a matrix; for an example of a data matrix see Section 5.6.1, Table 5.1. The transpose of a matrix will be denoted by X and its inverse by −1 X . Vectors will be considered column vectors and will be represented with lower case bold letters, for example, x. A row vector is the transpose of the column vector and will be represented by x . Table 5.1 Example of a data matrix.

Sample A-lot 60 Sample D-lot 16 Sample D-lot 16 Sample C-lot 25 Sample B-lot 13 Sample C-lot 26 Sample C-lot 55 Sample A-lot 25 Sample D-lot 11 Sample B-lot 07 Sample A-lot 56 Sample B-lot 31 Sample C-lot 09 Sample D-lot 12 Sample D-lot 11 Sample A-lot 16 Sample B-lot 20 Sample C-lot 03 Sample A-lot 34 Sample B-lot 43 Sample B-lot 21 Sample B-lot 28 Sample C-lot 05 Sample A-lot 51 Sample C-lot 15 Sample C-lot 25 Sample C-lot 16 Sample D-lot 10 Sample D-lot 12 Sample D-lot 11 Sample A-lot 55

46

47

48

...

147

148

149

200428 198508 171930 197901 233630 198790 198754 200667 163494 140608 222919 156297 201904 188802 215380 202342 221278 203905 204017 202252 187675 171986 200235 205692 226108 210537 199902 180366 203134 155058 203778

5256 2237 685 4310 4764 4503 3764 7910 164 2731 3833 3035 4114 1727 3279 7188 3825 4016 6466 4157 3642 3338 3135 5744 4546 3639 4212 1206 2790 357 3812

878 666 1077 454 659 252 330 913 1160 3187 637 2722 52 911 500 886 337 230 860 1589 1792 2257 548 833 894 348 253 994 1069 1243 825

... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...

90514 143773 143161 15650 64011 15785 15842 78827 143163 57771 90806 58711 15984 143158 143770 81546 61270 16151 84266 62131 60591 59651 18463 86985 12886 19080 15817 143160 141922 143164 95953

45304 75155 71811 2494 33155 2125 2245 39446 70707 23885 46469 25344 2460 74020 77364 40887 30149 2443 42329 30237 28262 26803 2233 43770 2492 2628 2477 72916 76168 69602 48187

29760 39880 38764 517 17499 587 486 24862 38233 9080 29158 10438 640 39827 40943 25946 15320 701 27030 14784 13153 11795 1011 28114 480 1441 578 39296 41848 37701 31928

94


One of the major concerns in experimental design is choosing an adequate number of samples in relation to the number of variables. Some of the multivariate analyses that are described here, have their roots in social sciences such as psychology or sociology where many objects or subjects are measured with very few variables. This is one of the biggest differences between chemical data where measurements on one sample can result in hundreds of variables; for example, a mass spectrum could contain ion abundance measurements for 300 different m/z values for a single compound, IR spectroscopy might have response measured at 4000 wavelengths, and a chromatographic analysis, run for an hour at 10 Hz, would have 30,000 data points. Bender et al. [3] and Kowalski et al. [4] have studied the difficulty of finding the adequate number of samples in relation to the number of variables. They measured the reliability of data-sets by calculating the ratio of samples to variables (R = n/p). A ratio of R > 3 is adequate for pattern recognition analysis. This ratio is difficult to achieve with hyphenated methods such as GC/MS and a central problem of applying chemometrics to analytical data is that of reducing the number of variables to a more manageable level. Goodner et al. [5] recommended using as few variables as possible to develop chemometric models. In their work, they suggest a ratio of 6:1 to 10:1 samples to variables. Caution should be taken when selecting the samples for any training data-set; the samples should be independent of each other. Particular attention should be placed on obtaining samples that represent the entire population, for example, samples from different lots of samples, samples obtained with different raw materials, samples from different crop years, and so on. Consideration should also be given to collection of sufficient data to use in validating any multivariate models made from the training set. Once the data has been tabulated as in Equation (5.1), the next step is to examine the raw data and preprocess it. The data should be then examined visually, preferably in both tabular and graphical form. Viewing plots of the raw data may expose unusual values which may derive from transposition errors, equipment malfunction, or simply aberrant samples. Preprocessing is defined as ‘any manipulation of the data prior to the primary analysis’ [6]. Preprocessing can be performed on the samples (rows of X) or on the variables (columns of X). Examples of sample-preprocessing techniques include normalization, weighting, smoothing, and baseline corrections. These techniques place samples on the same scale (normalization), increase the influence of some samples over others (weighting), reduce the amount of noise

95


(smoothing), or reduce systematic variation (baseline correction). For instance, taking second derivatives of spectra removes any baseline offset and facilitates fair comparison of samples [2]. Normalization removes possible variation due to sample size; it is necessary when the detector signal is a function of sample mass as for most gas chromatographs detectors. Normalization to unit area is accomplished by dividing each element in the rows of X by the ‘1-norm’ (in chromatography, known as area % normalization): 1-norm =

p

|xj |

(5.2)

j=1

Normalization to unit length is accomplished by dividing every element in the rows of X by the ‘2-norm’:

p

x2j (5.3) 2-norm = j=1

Johansson et al. [7] proposed selective normalization as a way to evade the problems occurring from closure of the data. Sahota and Morgan [8] discuss selective normalization for chromatographic data and demonstrate the effects of normalization on correlated variables. Variable-preprocessing tools include mean centering and variable weighting. Mean centering is achieved by subtracting the mean of the variable vector from all the columns of X. Mean centering accounts for the intercept in a calibration model and is usually recommended prior to performing principal component analysis (PCA). Variable weighting tools include variable selection, variable scaling, and autoscaling. Variable selection is a vast field and many different algorithms have been developed for the selection of optimal discriminating variables; sometimes this is done to maintain an adequate sample to variable ratio. Variable scaling is also used to remove differences in units between variables, which can be accomplished by dividing each element of X (xij ) by the standard deviation of that variable (sj ) [2]. Application of both variable-mean-centering and scaling is known as autoscaling. Autoscaling both accounts for the intercept in a calibration model and removes scaling differences between variables. The elements of the autoscaled data matrix Z are obtained by: zij =

xij − xj sj

(5.4)

96


where

n 1 xj = xij n

(5.5)

i=1

is the mean for variable j, and ⎧ ⎫1 n ⎪ ⎪2 2 ⎪ (xij − xj ) ⎪ ⎪ ⎪ ⎨ ⎬ i=1 sj = ⎪ ⎪ (n − 1) ⎪ ⎪ ⎪ ⎪ ⎩ ⎭

(5.6)

is the standard deviation of variable j. Autoscaling is also commonly referred to as ‘standardizing’ each variable. Once autoscaling has been performed, each variable (column) in the transformed data matrix will have a mean of zero and unit variance. The selection of which preprocessing technique to use is important because the magnitude of the variance of a variable will determine its importance in any final model, that is, any approximation of the original data set. Sanchez et al. [9] investigated the effects of eight different preprocessing techniques prior to PCA on chromatograms and spectra obtained with high-performance liquid chromatography coupled to diode-array detection. The three best results were obtained (a) without any preprocessing, (b) with selective normalization, and (c) with log transformations. There are many more preprocessing algorithms, some more applicable to specific techniques, such as first and second derivatives for spectrometry and smoothing transforms based on Savitzki–Golay polynomial filter. Something to keep in mind is that transmittance does not vary linearly with concentration; therefore transmittance values should be converted to absorbance if a regression model is needed. Some software companies make a clear distinction when the preprocessing is applied to the rows (independent variables) or columns (specific to a set of samples) of the data-set, then the order on how they are applied is also defined. It is important for the user of any chemometric software package to be well informed on how the preprocessing is carried out; since there is no universal approach to carrying out preprocessing, different packages do them in different ways. In summary: • mean centering shifts the origin without altering inter-sample relationships, usually performed on data prior to a principal component analysis;


97

• autoscaling involves sequential application of mean centering, then variance scaling. Autoscaled variables have mean 0 and variance of 1. • derivatives/smooth are based on the Savitzki– Golay polynomial filter; these are ways to remove baseline features from spectroscopic or chromatographic data. • rule of thumb: transform as little as possible and avoid random transforms. Pattern recognition has been defined by Albano et al. [10] as: ‘a methodology for finding rules of classification, that is, given a number of classes, each of which is defined by a set of objects (the training reference sets) and the values of M measurements made on each of the objects, rules are defined that make it possible to classify new objects (the test set) on the basis of the same M measurements made on these new objects.’ Two different categories exist for pattern recognition: unsupervised learning and supervised learning. Unsupervised learning refers to analysis of data in an exploratory mode. The real identity of the sample is not included in the model development. The goal of unsupervised learning is to find relations and similarities between samples. For instance, given a number of chromatograms one might need to group them according to how similar they are. Examples of unsupervised learning techniques include principal component analysis (PCA) and hierarchical cluster analysis (HCA). Models developed by supervised learning techniques take into account the identity of the samples being investigated. The data matrix is increased with one more column (see Section 5.6.4, Figure 5.6). Optimal models are then built based on misclassification rates of classification algorithms. The lower the misclassification rate, the better the model. Examples of supervised techniques include k-nearest neighbors (k-NN), soft independent modeling of class analogy (SIMCA), and linear discriminant analysis (LDA).

5.2 HIERARCHICAL CLUSTER ANALYSIS (HCA) This is an unsupervised technique that displays in a dendogram the distances between samples. A dendogram is a two-dimensional plot which shows the distances between samples (see Section 5.6.3, Figure 5.2). When the distances are small, the samples are similar, when the distances between samples are large, samples are dissimilar. Since there are no pre-existing categories assigned to the samples, this type of analysis emphasizes the natural grouping in the data-set.

98


The most commonly used multivariate distance is the Euclidean distance, which is defined as: m 2 = (xaj − xbj )2 (5.7) dab j

where dab is the multivariate distance between samples a and b, and xaj and xbj are the sample vectors. In a HCA the distances between each pair of samples are calculated and samples are linked according to form clusters. The similarity measure is calculated as: dab similarityab = 1 − (5.8) dmax where dmax is the largest distance in the data-set. Similarity = 1 for identical samples and similarity = 0 for the most dissimilar samples. There are several ways to link samples: single link links to the nearest neighbor (that is, the sample with the smallest distance), complete link links to the farthest neighbor and center links to the centroid, median, or group average. When working with very different groups, any linkage will work; for poorly separated groups a centroid-based method (for example, group average or incremental) is recommended.

5.3 PRINCIPAL COMPONENT ANALYSIS (PCA) Pearson first described principal component analysis 100 years ago [11]. At that time, he did not propose a practical method for the solution of problems with more than two variables. Hotelling later expanded Pearson’s idea to a more practical method for calculation of principal components in 1933 [12]. Recent advances in technology; specifically data storage, retrieval, and increased processor speed make PCA a practical technique, even for large data-sets. Principal component analysis (PCA) builds linear multivariate models for complex data-sets [13–15]. PCA facilitates the simultaneous interpretation of the entire data-set by finding linear combinations of variables (or principal components, PCs) that successively account for the maximum variability of the original data-set. One of the goals of PCA is to exclude PCs that correspond to noise and keep only PCs associated with systematic variance, thus reducing the dimensionality of the data. Normally, PCA is performed on the covariance matrix. For an Xn×p the covariance matrix, C, is defined as: Cn×n = (Xn×p Xn×p )/(n − 1)

(5.9)


99

PCA is efficiently computed using singular value decomposition (SVD) [16]. The basis of SVD is that a matrix Xn×p can be expressed as the product of three matrices: Xn×p = Un×r Sr×r V r×p

(5.10)

where the matrix S is diagonal and the sk elements are the square roots of the non-zero eigenvalues of the square matrix XX . The columns of U are the eigenvectors of XX and the rows of V are the eigenvectors of X X. Wold et al. [14] have recommended PCA as ‘an initial step to any multivariate analysis to obtain a first look at the structure of the data, to help identify outliers, delineate classes. . .’. When a PCA is carried out, several diagnostic plots are obtained. These plots provide information on variables, the samples, and the model. The natural grouping between samples is the scores plot. Because PCA concentrates information related to variance in the first principal components, sample similarities and differences are emphasized in plots of the sample positions on these early factors. Sample points in these plots are called scores plots, and these scores plots are key to visualizing the sample relationships. Variables that are important in the matrix decomposition are more heavily ‘‘loaded’’ in the early principal components, thus these vectors are often referred to as loadings. The amount of variance captured during the decomposition, known as eigenvalues, will decrease as more components are captured. When these are plotted as a function of factor number (called a scree plot), the pattern of eigenvalues can be used to decide the number of PCs, that is, which are relevant (information) and which are irrelevant (noise). In summary, exploratory analysis, PCA, and HCA provide tools for observing natural differences among samples, for detecting unusual samples (outliers) and discriminating important variables by inspection of their PC loadings.

5.4 CLASSIFICATION MODELS These techniques are a two step process: build and validate. The goal of this type of analysis is to compare new samples against a previouslyanalyzed data-set. These types of methods are supervised methods, a class is assigned to each sample in the training set (the training set is used to build the model), then a class is predicted and assigned to samples with unknown classes. There are several types of classification models, but two commonly used are KNN and SIMCA.

100


5.4.1 k-Nearest Neighbors (k-NN) This is a classification method based on a distance comparison between samples. Multidimensional distances between samples are calculated, and then the predicted class of unknown samples is determined based on the identity of those samples closest to the unknown. The unknown identity is found by the number of samples closest to the unknown, and these are referred as ‘votes’. The number of optimal votes is the ‘k’ in the model’s name. It is important to keep in mind that the number of maximum votes (k) is restricted by the category in a data-set with the fewest number of samples. This model is based on the multivariate distance, as shown in the HCA section; the Euclidean distance is the distance between objects in a multidimensional space. The unknowns are classified to one and only one class and the quality of the assignment is unspecified. The unknowns are classified based on their proximity to samples already placed in categories. The predicted class of the unknown depends on the class of its k nearest neighbors. ‘Each of the k closest training samples votes once for its class, the unknown is classified with the class with most votes. The number of neighbors can range from one to one less than the training set. If during the prediction of an unknown sample, two classes received equal votes, the tie is broken by calculating the accumulative distances. The class with the smallest accumulative distance is chosen. Rule of thumb: while running k-NN, choose the maximum number of neighbors less than twice the size of the smallest category. The goodness value is used to measure the quality of the prediction for unknown samples and it is defined as: gi =

di − d q sd(dq )

(5.11)

• it is a value similar to the t value in statistics; it indicates the number of standard deviations units the unknown is from the average class distance; • negative values are normal; this means the sample belongs to the class with a high degree of confidence; • large values indicate that the sample has a poor fit.

5.4.2 Soft Independent Modeling of Class Analogy (SIMCA) This is another classification method that is based on principal component analysis. The basis of this model is to create principal components


101

for each separate class present in the data-set, followed by the selection of relevant principal components for each class, better known as ‘rank’. Once the rank has been determined for each class, ‘multidimensional boxes’ are constructed around each class. These boxes then define boundary regions for each class. Just like any other chemometric model, the training set is used to create it and the testing set validates/rejects its prediction ability. Another important advantage of using SIMCA as a classification model is its ability of determining if a sample does belong to any predefined category but SIMCA also determines if a sample does not belong to any class at all. k-NN will predict a class for an unknown sample regardless if the prediction is reasonable or not.

5.5 PRINCIPAL COMPONENT REGRESSION Prediction of the quantitative composition of mixtures requires calibration experiments that relate the instrumental measurements to the components of interest. The initial step for any calibration requires measurements for reference samples, also commonly referred to as the training or calibration set from which the calibration model is developed. In the second step, measurements for a training sample are obtained and related to the calibration model developed from the calibration data. This prediction step is normally repeated many times with new test samples. Principal component regression (PCR) [17, 18] is a multivariate calibration technique that consists of two steps. First, PCA is performed on the predictor matrix (X); secondly, the number and identity of the PCs to be included in the model is chosen. Selection of PCs based solely on the order of variance typically results in poor PCR models [17, 18]. It is much better to select PCs on the basis of their correlations with y (predicted vector). If there is high correlation between a PC and the dependent variable vector y, then the PC is a good predictor and should be included in the regression. Regression coefficients are calculated in PCR with: U y bp×1 = Vp×k S−1 k×k k×n n×1

(5.12)

To validate a PCR model the data is split into two sets: training and testing. The model is built with the training set, and the testing set is used for prediction. Cross-validation (CV) is a diagnostic approach widely used in chemical applications because of the limited size of data-sets. CV makes use of all the cases and it gives more useful information regarding

102


the stability of the tree-structure. Parsimonious treatment of the data-set with cross-validation (CV) is used when there are not enough samples for separate training and testing sets. In this case, a portion of the data is left out at a time and the prediction is obtained using a model developed with remaining data. In v-fold cross-validation the original sample is divided at random into v subsets, each containing the same number of cases. When a single sample is left out this approach is also called jackknifing. CV proceeds as follows: the data is divided into v random portions and then the first portion is reserved for testing. A model is obtained with the remaining v-1 (training set) portions and tested using the first portion (test set) that was left aside. In the next step, the data is rotated and a second portion is reserved for testing. A second model is then obtained excluding the second portion and including the first portion. Then the second portion is used to measure the error rate of the model. This process is repeated until all the portions of the data have been rotated through testing and a cross-validated error of prediction is reported.

5.6 EXAMPLE OF DATA ANALYSIS FOR CLASSIFICATION MODELS 5.6.1 Tabulating Data The first step to any multivariate analysis is arranging the data in a spreadsheet. This data matrix will have the variables arranged in columns and the samples in rows. If the data set is chromatographic, variables can be area peaks at different retention times, if the data is spectra, then the variables can be absorbance at different wavelengths. If the detector is a mass spectrometer, the data can be arranged as variables being ion counts. It is important to have a fast method that arranges the data in a spreadsheet and some companies specialize in macros that create this data matrix. For this example, a data matrix with four sample types (A, B, C and D) and 105 variables was chosen. Notice that even though, the data has only four samples, the samples have many replicas. An important factor to keep in mind is that the chemometric model will be as good as the data collected. The samples should be independent samples that represent their population variance. For example, samples from different lots, samples collected at different


103

times, samples using different raw materials. The replicas should not be the same exact sample from a single lot which has been run several times; this will lead to not capturing the variance of the category that is being modeled. Table 5.1 shows the example data.

5.6.2 Examining Data This step should be done without any preprocessing and with preprocessing. There are several ways of preprocessing a data matrix, on their variables, on the samples or on any combination of both: • on variables – columns: – mean centering, – range scale, variance scale, autoscale: • on samples (transforms) – rows: – normalization, – weighting, – smoothing, and – baseline corrections. Once the data is in a tabulated format, the user must find multivariate software to analyze it and/or create models. For this example, we will use Pirouette [19]. The data can then be imported from a variety of formats, such as Excel, text file, and so on. A preliminary analysis should be carried out that determines if any sample looks extremely different or if there are errors in variables. Examination of the data can be done with plots using two variables (biplots). It can be seen by examining the bi-plot in Figure 5.1 that there is a sample that appears different than the others. This sample could be an outlier. If the task of comparing bi-plots is too complex (too many variables), you can move into multivariate exploratory analysis.

5.6.3 Multivariate Exploratory Analysis For this analysis, the samples are explored without a prior assignment of classes. The goal is to determine if they are natural groupings or if there is any correlation or structures among samples and variables. Figure 5.2 shows an HCA dendrogram. In this example, there are four clusters well separated. This indicates that the four categories, A, B, C,

104

PRACTICAL ANALYSIS OF FLAVOR AND FRAGRANCE MATERIALS 12

Variable 2

10 8 6 4 2 0

5

0

10

15 Variable 1

20

25

Figure 5.1 Biplot of variables 1 and 2. Sample A Sample A Sample A Sample A Sample A Sample A Sample A Sample A Sample B Sample B Sample B Sample B Sample B Sample B Sample B Sample C Sample C Sample C Sample C Sample C Sample C Sample C Sample C Sample D Sample D Sample D Sample D Sample D Sample D Sample D Sample D 1.0

Cluster A

Cluster B

Cluster C

Cluster D

0.5 Similarity

0

Figure 5.2 HCA of data-set with four categories.

and D have inheriting natural separation and this set could be used for a classification model. The most commonly used multivariate analysis is PCA. There are a lot of diagnostic plots when this model is created. An example is shown in Figure 5.3 shows a scree plot (see section 5.4). The selection of the

105

MULTIVARIATE TECHNIQUES 20 000.00

16 000.00

12 000.00

8000.00

4000.00

0.00 0

1

2

4

3

5

6

Figure 5.3 PCA scree plot.

number of PCs to keep is important as discussed previously. In this example, it appears that three PCs capture enough variance. Another PCA plot, and probably the one most commonly recognized, is the scores plot. An example is shown in Figure 5.4. This plot has axes created in the order of amount of variance in the data; each PC is orthogonal to the other. In Figure 5.4, the projection of samples into the first three PCs is displayed. It is important to report which PCs are displayed (in this case, the first three) and the amount of variance captured (in this case 98.65%). Factor2

Factor3

Factor1

Figure 5.4 PCA scores plots, ellipsoids do not represent confidence regions (98.65% variance captured within the first three principal components).

106


93 Loadings

69

105 107 11 9

0.0 57

13 3

81

–0.5

46 Variables

Figure 5.5 PCA Loadings plots.

In some research projects, the chemometric analysis goes as far as showing a scores plot (finding natural grouping). This is a problem if the data is to be used as a classification tool. It can also be a problem if the variance of the population was not properly captured. If for example, all samples of one category were analyzed on a specific day, then a PC may have captured variance due to the day and not due to the sample type. Randomization of the order in which samples are run for a training set is highly recommended to avoid capturing systematic errors. Another important plot that should be examined while carrying out a PCA analysis is the loadings plot for all the chosen PCs. Figure 5.5 shows the loadings for the first principal component. This plot shows what variables were important for the creation of the first PC. In this figure, variables 46 and 93 appear to be important. If these were ion masses, then maybe a compound with those ions could be differentiating the samples. The loadings plots for PC 2 and PC 3 are not shown here but should also be examined since the scree plot indicated three PCs were adequate to keep for this set.

5.6.4 Creation of a Classification Model with a Training Set and Validation with a Testing Set For this type of analysis, we need to add one more column to the data matrix: each sample class assignment. The training set becomes as shown in Figure 5.6. For k-NN the number of nearest neighbors (k) is chosen with the training set; k is made as large as possible but the number of misses is kept small. Once this is done the testing data-set is evaluated. As seen

107


Variable Sample C Sample A Sample B Sample C Sample D Sample A Sample B Sample C Sample D Sample A Sample B Sample C Sample D

46 226108 203778 202252 203134 210537 222919 233630 215380 198754 200428 221278 198508 200235

47 4546 3812 4517 2790 3639 3833 4764 3279 3764 5256 3825 2237 3135

48 894 825 1589 1069 348 637 659 500 330 878 337 666 548

147 12886 95953 62131 141922 19080 90806 64011 143770 15842 90514 61270 143773 18463

148 2492 48187 30237 76168 2628 46469 33155 77364 2245 45304 30149 75155 2233

149 480 31928 14784 41848 1441 29158 17499 40943 486 29760 15320 39880 1011

class

3 1 2 3 4 1 2 3 4 1 2 3 4

31 ´ 105

31 ´ 1

Figure 5.6 Data matrix before classification models, with class column added. Table 5.2 k-NN classification of unknown samples using goodness value. Class 1 Unknown 1 Unknown 2 Unknown 3 Unknown 4 Unknown 5 Unknown 6 Unknown 7 Unknown 8 Unknown 9 Unknown 10 Unknown 11 Unknown 12 Unknown 13

Class 2

Class 3

Class 4

−0.58756 47.93282 61.98627 31.5728 51.801 0.632427 51.19658 31.9879 178.6284 169.7851 289.4688 1.003831 −0.52808 46.51608 63.17764 31.07885 50.2593 0.347203 56.22467 30.8833 46.96932 32.74984 −0.3749 36.09097 163.0246 155.9081 268.1997 −1.22517 −0.38146 47.05276 67.22506 29.8066 53.20723 0.563078 47.55336 30.69936 46.73367 33.31105 −0.11794 35.9984 160.5863 153.6007 265.2962 −1.92559 −0.07051 47.53406 67.21384 30.61489 50.82423 −1.22794 50.84089 29.97689

in Table 5.2, goodness values can be negative. This means the sample belongs to the class with a high degree of confidence; in Table 5.2 the class with the lowest goodness value is underlined. The large goodness values indicate that the sample has a poor fit. SIMCA models develop PCA models for each category. A multidimensional ‘box’ is created for each class and the unknown samples are classified according to which box (if any) they belong to. Prediction of unknowns: combination of how far the sample is from (a) the PCA model (i.e., residual), and (b) its projection is from the SIMCA box boundary. The boundaries depend on the number of samples of the training set. (Rule of thumb: have at least 10 independent replicas per class.) These boundaries also depend on the critical value selected

108


(probability cut-off); p = 0.95 will predict differently from p = 0.999. The box drawn has straight edges but in reality the region in space is curved due to the nature of the F test. In order to optimize the model, the optimal number of principal components for each category must be selected. The number of principal components can be different for each class. Class separation can be improved by selecting variables and examining total modeling power and discrimination power. Another important diagnostic to examine for a SIMCA model is the interclass distance. This measurement indicates how well the classes are separated from each other. As a good rule of thumb, interclass distances greater than three are considered well separated. For the samples displayed in Table 5.3 these distances indicate good separation between samples. SIMCA develops principal component models for each category of the training set. The bounding ellipses in Figure 5.7 form a 95 % confidence interval for the distribution of these categories. In this case, Table 5.3 Interclass distances for a SIMCA model. Interclass distances

CS1 CS2 CS3 CS4

CS1@4

CS2@2

CS3@2

CS4@2

0 18.99593 21.87419 17.42387

18.99593 0 11.42804 23.16223

21.8742 11.42804 0 22.9642

17.42387 23.16223 22.96419 0

Factor2

Factor3

Factor1

Figure 5.7 SIMCA scores plot, ellipsoids represent 95 % confidence regions.


109

the projection of the variables indicates good clustering between samples without overlap. Another indication of a good SIMCA model is the interclass distances between samples. Once classification models have been created and validated, they should be tested periodically. For example, some of the samples used in the creation or validation of the model should be saved. Each time a sequence of unknown samples is run, some of the saved samples should be run alongside. If the saved samples predict accordingly then the model is still valid, if not close inspection to the model should be carried out to update or re-create the model. This chapter represents a small introduction in chemometrics. For comprehensive information regarding multivariate analysis, please see reference 5. Beeve, Pell and Seasholtz [5] advocate six habits for a good chemometrician. These six habits include not only examining data, preprocessing data, estimating models as shown in this chapter but also include validating the model, using the model for prediction and validation of predictions.

REFERENCES 1. Frank, I. E. and Kowalski, B. R. (1982) Chemometrics. Anal. Chem., 54, 232R– 243R. 2. Kinton, V. (2001) Chemometric techniques for modeling and classification of composition and identity in multivariate analytical chemical data. Ph. D. Thesis, University of South Carolina, Columbia, SC, USA. 3. Bender, C. F., Shepherd, H. D., and Kowalski, B. R. (1973) Alternate representation of mass spectra for the spectral identification problem (pattern recognition). Anal. Chem., 45, 617– 618. 4. Kowalski, B. R. and Bender, C. F. (1972) Pattern recognition. A powerful approach to interpreting chemical data. J. Am. Chem. Soc., 94, 5632– 5639. 5. Beebe, K. R., Pell, R. J., and Seasholtz, M. B. (1998) Chemometrics: A Practical Guide, John Wiley & Sons, Inc., New York, USA. 6. Goodner, K. L., Dreher, J. G., and Rouseff, R. L. (2001) The dangers of creating false classifications due to noise in electronic nose and similar multivariate analysis. Sens. Actu. B, 80, 261–266. ¨ 7. Johansson, E., Wold, S., and Sjodin, K. (1984) Minimizing effects of closure on analytical chemistry. Anal. Chem., 56, 1685– 1688. 8. Sahota, R. S. and Morgan, S. L. (1992) Recognition of chemical markers in chromatographic data by an individual feature reliability approach to classification. Anal. Chem., 64, 2383– 2392. 9. Sanchez, F. C., Lewi, P. J., and Massart, D. L. (1994) Effect of different preprocessing methods for principal component analysis applied to the composition of mixtures: detection of impurities in HPLC-DAD. Chemomet. Intell. Lab. Sys., 25, 157– 177. 10. Albano, C. et al. (1978) Four levels of pattern recognition. Anal. Chim. Acta, 103, 429– 443.

110


11. Pearson, K. (1901) On lines and planes of closest fit to systems of points in space. Phil. Mag., 2, 559–572. 12. Hotelling, H. (1933) Analysis of a complex of statistical variables into principal components. J. Ed. Psych., 24, 417– 441 and 498–520. 13. Jackson, J. E. (1980) Principal components and factor analysis: Part I-principal components. J. Qual. Tech., 12, 201–213. 14. Wold, S., Esbensen, K., and Geladi, P. (1987) Principal component analysis. Chemomet. Intell. Lab. Sys., 2, 37–52. 15. Jollife, I. T. (1986) Principal Component Analysis, Springer Verlag, New York, USA. 16. Press, W. H., Flannery, R. P., Teukolsky, S. A., and Vetterling, W. T. (1986) In Numerical Recipes: The Art of Scientific Computing, Cambridge University Press, Cambridge, UK, pp 60–72. 17. Egan, W. J., Brewer, W. E., and Morgan, S. L. (1999) Measurement of carboxyhemoglobin in forensic blood samples using UV-visible spectrometry and improved principal component regression, Appl. Spectrosc., 53, 218– 225. 18. Martens H. and Næs, T. (1989) Multivariate Calibration, John Wiley & Sons, Inc., New York, USA. 19. Pirouette (2007) Version 3.12, Infometrix, Seattle, USA.

6 Electronic Nose Technology and Applications Marion Bonnefille Alpha MOS, Toulouse, France

6.1 INTRODUCTION Over the last decade, ‘electronic sensing’ – or ‘e-sensing’ – technologies have undergone important developments from a technical and commercial point of view. The expression ‘electronic sensing’ refers to the capability of reproducing human senses using sensor arrays and pattern recognition systems. The first sense to be reproduced by an instrument was hearing and devices that could be called ‘electronic ears’ were developed for industrial purposes. In recent years, ‘electronic eyes’ have also experienced a number of developments for biometric applications such as iris recognition for safety purposes or for industrial routine quality control. For the last 15 years, research has been conducted to develop technologies, commonly referred to as electronic noses, which could detect and recognize odors and flavors. The stages of the recognition process are similar to human olfaction and are performed for identification, comparison, quantification, and other applications. However, hedonic Practical Analysis of Flavor and Fragrance Materials, First Edition. Edited by Kevin Goodner and Russell Rouseff. © 2011 Blackwell Publishing Ltd. Published 2011 by Blackwell Publishing Ltd.

112


evaluation is a specificity of the human nose given that it is related to subjective opinions. These electronic nose devices have undergone much development and are now used to fulfill industrial needs in research and development departments, quality control for flavor and fragrance, food and beverages, packaging, pharmaceutical, cosmetic and perfumes, and chemical companies. More recently they can also address public concerns in terms of olfactive nuisance monitoring with networks of on-field devices. Most electronic noses use sensor-arrays that react to volatile compounds on contact: the adsorption of volatile compounds on the sensor surface causes a physical change of the sensor. A specific response is recorded by the electronic interface transforming the signal into a digital value. Recorded data are then computed based on statistical models. The more commonly used sensors include metal oxide semiconductors (MOS), conducting polymers (CP), quartz crystal microbalance, surface acoustic wave (SAW), and field effect transistors (MOSFET). Recently, other types of electronic noses have been developed that utilize mass spectrometry or ultra fast gas chromatography. However, this chapter will only focus on sensor-array based technologies and will present: • • • • •

sampling systems, detection technologies, data treatment tools, application range, case studies.

6.2 HUMAN SMELL AND ELECTRONIC NOSES The sense of smell in humans is one of the most important senses and the one that historically is useful for identification and protection. The human process of olfaction includes: • the perception of a flavor by odor receptors, • the conversion of the perception into a signal, • the transduction of the signal to the limbic system of the brain, that is, the area in which emotional responses occur and which is associated both with memory and with physiological responses independent of conscious decisions, and • passing the signal on to the neocortex for labelling (recognition, etc.).

113

ELECTRONIC NOSE TECHNOLOGY AND APPLICATIONS

Aged apple flavor

Human sensing process Odor receptors

Analysis by brain

1

2

3 and 4

Acquisition

Data treatment

Comparison and decision

Gas sensors detection

Analysis by software

Fingerprint recognition

Fingerprint recognition

Instrumental process

Figure 6.1 Comparison of human sensing and instrumental sensing of aroma volatiles.

The electronic nose was developed in order to mimic human olfaction which functions in an unseparated mode: that is, an odor/flavor is perceived as a global fingerprint (see Figure 6.1).

6.3 TECHNIQUES TO ANALYZE ODORS/FLAVORS In industry, aroma assessment is usually performed by human sensory analysis or by gas chromatography (GC, GC/MS). The latter technique gives information about volatile organic compounds but the correlation between analytical results and global odor perception is not direct due to potential interactions between several odorous components. This is why electronic noses are being used more widely in industrial applications.

6.3.1 Sensory Panel Companies look to non-trained sensory panels for subjective evaluations (for instance consumer tests). To evaluate the conformity of a manufacture’s product, or to assess the quality or origin, they often have to constantly train panelists and establish standardized quotation scales or criteria. Trained sensory panels can detect aroma differences in a sample,

114


from major differences to minor ones. Sensory evaluation is certainly the most widespread among odor/flavor assessment techniques. More and more sensory panel analyses are being allocated to new product development (research and development) or to starting (as a reference) and ending (end-user evaluation) points of a production process. The two chief advantages of human sensory panels are the human sensitivity to some molecules (concentrations in the sub-ppb range) and the selectivity of the human nose. It is the most common method used by consumers for the final assessment of a product. However, sensory panels present some major drawbacks for industrial purposes: • • • •

limited availability, time consumption and costs, human fatigue which limits the number of analyses per day, related subjectivity and variability entailing variability in scientific evaluation, • reluctance in risking the testing of unpleasant or noxious products.

6.3.2 GC and GC/MS Analytical techniques such as gas chromatography (GC) and mass spectroscopy (MS) are also commonly used by chemists to identify compounds and their concentrations in an odor/flavor. Given that they are separation techniques, the results are not directly linked to information produced by human sensory panels and this correlation is difficult to achieve. They are perfectly suitable for the analysis of a single pure molecule but it is much more difficult to interpret results for odor/flavors that consist of complex mixtures of compounds.

6.3.3 GC/Olfactometry A method combining GC with human evaluation (GC/olfactometry) is sometimes used as a complement: for example, an analyst can smell individual compounds of a mixture, after separation by a GC-column. By assessing the type and intensity of each component, this method gives detailed information on the composition of an odor. Nevertheless, it is not a global analysis that will enable comparison between various flavors. In addition it is a very lengthy technique to implement and therefore it is not suitable for routine quality control tests.


115

6.3.4 Electronic Nose As a first step, an electronic nose needs to be trained with qualified samples so as to build a database of reference. Then the instrument can recognize new samples by comparing volatile compound fingerprints to those contained in its database. Thus they can perform qualitative or quantitative analysis. The electronic nose presents various advantages for flavor measurement: • • • • • • • •

no or very little sample preparation, consistency and reproducibility of the measurement, rapidity of obtaining results, high throughput of analyses, permanent availability, nondestructive and global analysis of an odor/aroma, fingerprint results as human assessment, results correlated with human perception due to multivariate data treatment.

These advantages explain why electronic noses are used not only in research and development laboratories (as a fast and extensive screening technique) but also at the production stage (for rapid quality control). In the flavor and fragrance industry and related areas, the electronic nose proves to be a powerful tool which broadens aroma analysis capabilities. However, the electronic nose cannot replace the human nose for subjective analysis (for example, consumer preference tests or qualification of a new product never analyzed before). More generally, electronic nose analyzers can be applied to all types of volatile compounds, for example, to monitor batch to batch variation, raw material variability, and samples that are in conformity/out of specification for whatever reason.

6.3.5 Electronic Nose Technology and Instrumentation 6.3.5.1 Architecture Gardner and Bartlett (1993) defined the electronic nose as ‘an instrument including a set of electronic chemical sensors with a cross selectivity,

116


Sample delivery system

Detection system

Computing system

Figure 6.2 Block diagram of components of the process involved in electronic nose operation.

and a fitted pattern recognition system capable of recognizing simple or complex odors’. Figure 6.2 illustrates the three major parts that comprise an electronic nose: • a sample delivery system, • a detection system, • a computing system. Sample delivery system This enables the generation of the headspace (volatile compounds) of a sample, which is the fraction analyzed. The system then injects this headspace into the detection system of the electronic nose. The sample delivery system is essential to guarantee constant operating conditions. Detection system This consists of a sensor set, which is the ‘reactive’ part of the instrument. When in contact with volatile compounds, the sensors react, which means they experience a change of electrical properties. Each sensor is sensitive to all volatile molecules but each in their specific way. Computing system This works to combine the responses of all of the sensors, which represents the input for the data treatment. This part of the instrument performs global fingerprint analysis and provides results and representations that can be easily interpreted. Moreover, the electronic nose results can be correlated to those obtained from other techniques (sensory panel, GC, GC/MS).


117

Figure 6.3 Photo of the Alpha MOS Fox electronic nose with autosampler. Source: Alpha MOS, France.

Figure 6.3 represents a typical system commercially available on the market and consisting of: an autosampler, a sensor array based electronic nose including 18 metal oxide sensors (the FOX from Alpha MOS, France) and a computer. 6.3.5.2 Air Generator In the early days of electronic noses, ambient air changes during the day rapidly generated variations in sensor response over time. To overcome these variations linked to external conditions, electronic noses are now connected to a source of pure and constant air. In order to operate properly and in optimum conditions, it is recommended that electronic nose instruments are fuelled with a constant and high quality pure air as the carrier gas with the specifications: • • • •

H2 O < 5 ppm Cn Hm < 5 ppm O2 + N2 > 99.95 % O2 = 20 % +/− 1 % .

The source of pure air consists of a total organic compounds (TOC) generator which produces purified air from compressed air (from an air

118


compressor or compressed air delivery network). The air generator uses a combination of filtration, combustion, and pressure swing adsorption to remove hydrocarbons, CO2 , and water from the compressed air. 6.3.5.3 Sampling Electronic nose instruments are used for volatile organic compound and odor analyses. Sample preparation with an aim to generate headspace remains a key parameter to provide consistent and accurate results. Since the 1990s, several new headspace sampling techniques have been developed to adapt analytical instrument capabilities to all kinds of samples and extend their performance in terms of detection limits. Automated or semi-automated sampling devices At first, analyses were run by presenting the sample at the ‘entrance’ to the electronic nose. Since the reproducibility of the analyses was not conclusive, several developments have been implemented. For example Alpha MOS Company introduced a ‘measurement chamber’ to isolate the sample in a vial with air flowing directly on the content through valves. The vial was heated to generate a headspace, and air was sent to the sample manually after triggering data recording. In order to improve synchronization of air flow and data recording, a four-way valve was used. In the meantime, a mass flow controller (MFC) was installed to monitor a constant gas vector flow. Between the end of 1995 and the beginning of 1996, autosamplers were connected to the instruments. An autosampler consists of: • a tray to store several vials containing samples, • a multiposition oven with orbital stirring, which proves to be efficient to guarantee homogeneity, • a heated syringe block (various volumes available). The multiposition oven allows the heating and stirring of several samples simultaneously so as to overlap preparation times and thus reduce the overall analysis sequence time. At the present time the autosampler and its operating parameters can be piloted directly from the electronic nose software. The use of autosamplers provides automation and facilitates analyses while allowing a better understanding of issues related to sampling. Moreover, these devices guarantee the reproducibility of: heating/stirring times, stirring speed, heating accuracy, and injected volume.


119

Instead of the autosampler, other sampling devices with a lower degree of automation were developed to meet the needs of users who perform fewer analyses and who do not require a high throughput: • the 2T station allows the automation of the heating time of various samples simultaneously and ensures reproducible conditions (time, temperature); • the Matrix system also performs automated heating of samples and further enables semiautomated SPME extraction of headspace. SPME Among the different headspace sampling methods, solid phase microextraction (SPME) is very innovative and completely fulfills electronic nose requirements. In SPME, a coated fused silica fiber is introduced into the sample, and volatile organic compounds (VOCs) are adsorbed onto the coating. Then the analytes are desorbed from the fiber to the analyzer by a heated injection port. By changing the coating type or thickness, the selectivity can be altered in favor of more or less volatility. This property can be monitored to selectively adsorb sample components, as desired. Headspace SPME is ideal for: • minimizing the matrix background during an analysis, • revealing additional VOCs that have been obscured in a direct headspace technique, • improving the quantity of compounds to be analyzed and thus increasing the sensitivity, and • increasing the selectivity in headspace sampling. Thermodesorption devices Thermodesorption devices are aimed at performing enrichment of volatile or semivolatile compounds during headspace analysis. Their working principle is based on dynamic extraction of headspace and automated thermodesorption in the instrument. The sample headspace is repeatedly pumped by the syringe. Sensitivity can be improved by increasing the number of pumping strokes. During thermal desorption into the instrument, the adsorbing phase is rapidly flash heated in the injection port. Among recent sample concentration techniques, the solid phase dynamic extraction (SPDE) from Chromtech, Germany, consists of a syringe comprising a needle with an inner coating of adsorbing material. Another device commercially available, the ITEX from CTC, Switzerland) uses a syringe including a microtrap (Tenax or activated charcoal) placed between the syringe and the needle.

120


A dynamic thermodesorption device, the TDAS-2000 from Chromtech, Germany) was developed for fully automated thermal desorption to analyze organic compounds that vaporize or semivaporize in solid, liquid, or gaseous samples. It assumes that the volatiles have been previously trapped (during a dynamic extraction called purge) on a sorbent in a tube. The desorption is performed at high temperature. This technique allows the concentration of medium and low volatile compounds and their detection with the analytical system. This dynamic extraction is more efficient than a static one. Moreover this module can be mounted on commercially available autosamplers (such as CTC) so that the transport of the sample is over short distances, which yields results which are highly reproducible and even more sensitive. There are no critical parts involved, such as heated transfer lines and/or switching valves, so errors related to sample contamination and low reproducibility are eliminated. Other sampling options priate sampling:

Specific modules can be used to achieve appro-

• For temperature-sensitive samples that need to be stored at a low temperature, a Peltier cooler can be mounted on the autosampler to cool down samples on their tray. • For on-line analysis a flow cell can be used to aspirate the sample headspace at chosen intervals or to spike a liquid or gas stream at certain time intervals with a reagent or standard. The main applications concern on-line monitoring of processes (fermentation, chemical reactions, a pilot plant, and cooking processes) and environment analysis (waste or drinking water lines). • To automate the weighing-in of samples the Balance Pal option automatically weighs samples and transfers them to the detector, while transmitting the weight data to the computer. • For large volume samples that cannot be introduced into a standard vial (10–20 mL) the Baker oven from Chromtech, Germany) can contain samples up to 750 mL. This module is designed to generate headspace from large volume samples. • Measurement in microbial applications: Petri dishes with a specific sealing system were developed to directly sample and analyze volatile compounds emanating from cultures. This module can be directly fixed on the headspace sampler. Petri dish sampling can be temperature controlled using a cooling device. Once the inoculation of organisms is performed in the Petri dish, the growth rate can be directly monitored via the electronic nose measurement.


121

6.3.5.4 Detection Technologies Metal oxide sensors (MOS) As shown in Figure 6.4, metal oxide semiconducting sensors are generally made of: • a ceramic substrate, • metal electrodes to measure conductance, • a heating element (wire) to activate reactions and to allow the elimination of contaminants on the sensor surface, and • a coating consisting of a semiconducting metal oxide layer. The metal oxide coatings used are mainly of two types: • n-type (n = negative electron) oxides that include zinc oxide, tin dioxide, titanium dioxide, or iron (III) oxide; they are more sensitive to oxidizing compounds because the excitation of these sensors results in an excess amount of electrons in its conduction band. • p-type (p = positive hole) oxides such as nickel oxide or cobalt oxide; they are more reactive to reducing compounds since they develop an electron deficiency under excitation. Metal oxide sensors are known to have a low sensitivity to moisture and are quite sturdy. The typical operating temperature range is from 400 ◦ C to 600 ◦ C. Volatile compounds are adsorbed onto the surface of the semiconductors thus generating a change in the surface electrical resistance which is also a function of the gas concentration. Figure 6.5 describes the simplified mechanism commonly proposed, for instance by Morrison and Kohl, to explain sensor/gas interaction. Lead Metal electrodes Lead Ceramic substrate

Metal oxide layer Heater

Figure 6.4 Diagram of a metal oxide sensor.

122


O2

H2O OH− H+

Carrier gas 1

Gas: O2 H2O OH− H+

e− e− e− e−

R0 (Initial resistance)

H O2 2

Gas + Headspace 1

H2OSOH− H+

e− e− e− e− e− e−

R (Modified resistance) R0

R0

Adsorption

3

Desorption 2

Rmax

Figure 6.5 Representation of the resistance change which occurs on a metal oxide sensor when a volatile is absorbed on its surface.

• At equilibrium, under reproducible and constant carrier gas flow (usually pure air produced by a TOC generator), the sensor has a stable resistance R0 . • When sample headspace is injected with the carrier gas, the volatile compounds that form the headspace are adsorbed on the sensor surface and react with the oxygen contained in the carrier gas. The sensor resistance is thus modified up to an optimum value Rmax . These data (maximum resistances) are recorded over time by the system for further treatment. The presence of oxygen is absolutely necessary in this mechanism. Then desorption of volatile compounds occurs and the sensor recovers its inital resistance R0 , since the process is completely reversible. Sensitivity (See Table 6.1) To determine sensor sensitivity, various formulas are proposed in the literature. Generally, they imply values measured with air and with gas. The main parameters that influence the sensitivity of metal oxide sensors are: • • • •

the characteristics of the sensor material, the operating temperature of the sensor, the ambient conditions (humidity, temperature), and the composition and concentration of the gas.

123


Table 6.1 Some detection thresholds found in the literature. Compound

Detection threshold

Operating conditions

H2 S

0.1 ppm

Methanol, ethanol propanol and acetone

10 ppb to 100 ppb

NO

1 ppm

H2

0.1 ppm

SnO2 sensors Operating temperature: 300 ◦ C Commercial sensors (Figaro) TGS 812, TGS 824, and TGS 800; static injection in a volume of 20 L SnO2 sensors Measuring cell: 1700 cm3 Range of temperature: 25–300 ◦ C MOS-Pd sensors coupled with a Pt strand (catalytic action)

The most sophisticated electronic noses have included devices that control these parameters and thus achieve stable operating conditions (chambers with thermotats to control sensor temperature, a source of pure air to limit moisture and contamination, etc.). Conducting polymers sensors Conducting polymer sensors are composed of an organic conducting polymer film laid down on a silicon or carbon substrate including an electrode, as shown on Figure 6.6. The polymer is obtained through electrochemical polymerization of an aromatic monomer solution (example: pyrrole, thiophene, anilline, indole . . . ) with an electrolyte in a solvent. They can be used at room temperatures but not much higher. They react to many volatile compounds but they are more sensitive to moisture (water decreases resistance). GAS Conducting polymer

Electrode

Insulating layer Substrate

V

Figure 6.6

Diagram of a conducting polymer sensor.

124

PRACTICAL ANALYSIS OF FLAVOR AND FRAGRANCE MATERIALS Polymer film

Electrodes

Quartz crystal

Figure 6.7 Diagram of a quartz microbalance sensor.

When exposed to various volatile compounds, the polymer film conductivity is modified and electrical resistance variation is measured; conduction is achieved by electrons and not by ions. These changes are reversible at room temperature. For practical purposes, either the current intensity is measured under constant voltage or voltage is measured under constant intensity. Quartz microbalance sensors (QMB) Two types of sensors from this family of piezoelectric crystal sensors exist: those vibrating on a bulk acoustic mode (quartz microbalance) and those vibrating on a surface acoustic mode (SAW). As shown in Figure 6.7 these sensors consist of a quartz crystal on which are placed metal electrodes (aluminum or gold) coated with a polymer film. Usually the polymer films are hydrophobic (polyethylene, polystyrene, etc.). Acoustic waves are generated by an oscillating electrical field induction in the device. Surface acoustic wave (SAW) sensors These sensors use surface acoustic wave transmission, which is a normal and periodic alteration of the surface under alternative voltage. The solid is usually a piezoelectric crystal and the wave is induced by microelectrodes. In both types of sensors, the volatile compounds adsorbed on the film lead to a change in mass that modifies the wave spreading. The measurable parameter is the oscillating frequency variation. Table 6.2 presents the comparison of the characteristics of the three types of sensors previously described. Metal oxide semiconductors field effect transistor (MOSFET) sensors As shown in Figure 6.8, MOSFET sensors consist of three layers including a silicon semiconductor, a silicon oxide insulator, and a catalytic metal. The latter is commonly called the gate and is usually made of palladium, platinum, iridium, or rhodium.

125


Table 6.2 Comparison of metal oxide, conducting polymers & quartz microbalance sensors. Metal oxide sensors Conducting polymers Quartz microbalance Sensitivity Selectivity Stability Sample humidity effect Reaction mode Measurement function Ruggedness Sample injection temperature Resistance to poisoning/ damage

+++++ +++ +++++ Low

+ ++++ + Severe

+++ +++++ +++++ Low to mild

Oxidation Resistance change

Polarity Conductivity change

Polar or nonpolar Mass vs. frequency

+++++

Practical Analysis of Flavor and Fragrance Materials

Common Fragrance Flavor Materials

Common Fragrance and Flavor Materials

Flavor, Fragrance & Odor Analysis

Common Fragrance and Flavor Materials: Preparation, Properties and Uses

Perspectives in Flavor and Fragrance Research

Common Fragrance and Flavor Materials Fourth, Completely Revised Edition

Common Fragrance and Flavor Materials: Preparation, Properties and Uses, Third Edition

Common Fragrance and Flavor Materials: Preparation, Properties and Uses, Fifth Edition

Thermal Analysis of Materials

Thermal Analysis of Materials

Flavor Fragance and Odor Analysis Marsili

The encyclopedia of herbs: a comprehensive reference to herbs of flavor and fragrance

Thermal Analysis of Materials

Thermal Analysis of Materials

The encyclopedia of herbs: a comprehensive reference to herbs of flavor and fragrance

Handbook of flavor chracterization

Thermal Analysis of Polymeric Materials

Thermal Analysis of Rubbers and Rubbery Materials

Geotechnical materials measurement and analysis

Mechanics and Analysis of Composite Materials

Mechanics and Analysis of Composite Materials

Mechanics and Analysis of Composite Materials

The Flavor of Love

The Flavor of Love

Flavor chemistry and technology

Chemistry of Wine Flavor

The Flavor Of Love

Practical bifurcation and stability analysis

Flavor Perception

Practical Analysis of Flavor and Fragrance Materials

Common Fragrance Flavor Materials

Common Fragrance and Flavor Materials

Flavor, Fragrance & Odor Analysis

Common Fragrance and Flavor Materials: Preparation, Properties and Uses

Perspectives in Flavor and Fragrance Research

Common Fragrance and Flavor Materials Fourth, Completely Revised Edition

Common Fragrance and Flavor Materials: Preparation, Properties and Uses, Third Edition

Common Fragrance and Flavor Materials: Preparation, Properties and Uses, Fifth Edition

Thermal Analysis of Materials

Thermal Analysis of Materials

Flavor Fragance and Odor Analysis Marsili

The encyclopedia of herbs: a comprehensive reference to herbs of flavor and fragrance

Thermal Analysis of Materials

Thermal Analysis of Materials

The encyclopedia of herbs: a comprehensive reference to herbs of flavor and fragrance

Handbook of flavor chracterization

Thermal Analysis of Polymeric Materials

Thermal Analysis of Rubbers and Rubbery Materials

Geotechnical materials measurement and analysis

Mechanics and Analysis of Composite Materials

Mechanics and Analysis of Composite Materials

Mechanics and Analysis of Composite Materials

The Flavor of Love

The Flavor of Love

Flavor chemistry and technology

Chemistry of Wine Flavor

The Flavor Of Love

Practical bifurcation and stability analysis

Flavor Perception

Recommend Documents